On [Un]Churned, Brady Bluhm said that after using Staircase's MCP he expects people won't need to log into product UIs within about two years, because LLMs improve faster than other software.
Ann says Crunchbase launched its MCP server about a month before the recording and that the value lies in bringing data sources together through an AI interface that sits inside the seller's workflow. She says the choice of front-end AI tool, such as ChatGPT, Gemini or Grok, does not matter.
“it's about how you seamlessly bring together the data sources through the AI interface to deliver real value.”
Chuck Ganapathi says Gainsight launched its MCP interface about a month before Pulse, and all its products now have one. He says this lets customers work in their preferred agentic tool and use Gainsight without ever having to log into it, and that the Gainsight CLI lets admins configure the product through coding agents.
“you can use Gainsight without ever having to log into Gainsight”
Diane Wu says that in the old world CSMs used multiple tabs and tools to get the context they need, and she expects this to go away. She describes the LLM as the new workspace, with the interface being natural language rather than visual, at least today. She adds that this might change very quickly in the future.
“I'm a believer of the traditional application UI is going to eventually fade and go away and LLMS will become the new workspace”
Brady Bluhm says he had a product manager's existential moment after playing with Staircase's MCP and asking why anyone would log into his product in a year and a half or two years. He expects that change because LLMs themselves are improving faster than other software. He still builds for the UI today because not everyone works in LLMs all the time.
“in two years from now, I think that's going to change because the LLMS themselves evolved so much faster than any other product will too”
On [Un]Churned, Diego Ballona said Intercom built a Slack support feature for one customer in weeks, off-roadmap, then productized it to general availability in about three months.
Grant Clarke says Atlas records what reps do outside the agent's handoffs and classifies those actions into buckets. Forward deployed engineers, product managers and business analysts sit 'in the boat' with renewal managers to monitor this. When a category of escalation recurs multiple times, it becomes a product requirement. The FDE team develops it in maybe two weeks, tests it for a few weeks, and puts it back into Atlas.
“The FDE team can sprint, develop that in maybe two weeks, run a few weeks of testing, and then it's right back into Atlas to handle the next customer issue that comes up.”
Rick says a core offering may handle around 90% of what customers need, and that the remaining edge cases can be handled by building applications that complement the core product. He says this is more possible now, but that it adds complexity because forward-deployed engineering teams must work with core engineering. He says each organization should decide which things it will not build into the core product but will still say yes to.
“what are the things that we're not going to build into our core offering? But we should be able to say yes to”
Cassie says she is seeing a notable organizational change in portfolio companies: implementation, previously in the chief customer officer or post-sales lane, is moving into product. Companies view getting customers up and running as a product opportunity, and the best organizations productize what they learn from forward-deployed work immediately.
“implementation in many places is being moved out of the chief customer officer lane or the post-sales lane and into product”
A host says many AI companies have forward-deployed engineers but no plumbing connecting them to product, so the work is just implementation. He says the true FDE model is one where that work feeds back into what gets built, which is what made the Palantir model special. Cassie agrees, saying Primary has discussed the Palantir model internally and that Palantir built product off it rather than just selling services.
“The whole FTE concept was it was going to help Palantir figure out what to build and that feedback loop is what was so special about the whole model”
He said FDE works on early prototypes with specific customers. Once a product becomes a big part of the platform's infrastructure and needs more investment, a core product team takes ownership and productizes it for all customers.
“But at some point, it makes sense for a core product team to own the product because it becomes a big part of like our infrastructure of our platform.”
A customer had a large share of its support volume on Slack, which Fin did not support at the time. Engineers sat with the customer's head of support, and an engineer estimated a version could be built in two to three weeks. Intercom built it for that customer despite it not being on the roadmap, and because demand was validated, it was productized into a general availability product in about three months.
“we actually decided to invest in that because we validated demand.”
Diego said this is different from aggregated research or metrics, which he called very valuable but slower to act on. He said the FDE motion has led Intercom to build a better product for all customers and better internal tooling for its go-to-market team.
“when you get engineers sitting with customers directly, they feel the pain first hand.”
On Topline, Steve Cox cited a statistic that people use about 25% of their software tools, arguing more features create bloat rather than a better system.
Dan Lee said it is easy to build a ton of features, so the hard part is knowing what is important. Nooks lives by a do more with less value, and he described its approach as rigorously evaluating options against each other and setting clear north-star metrics and leading indicators.
“we can do anything but not everything.”
A host had observed that GTM platforms bolt features onto a left-hand menu without connected workflows. Keith agreed that everyone is racing to the roughly 100 features required to run a go-to-market team. He said making them performant and coherent together is really hard, and many players produce apps that in theory prospect, forecast and hold a CRM data model but aren't usable by a rep.
“create this Frankenstein app. You know, with a hundred features that in theory could prospect, in theory could forecast, in theory has a CRM data model, but is not a thing that a rep can use.”
He says companies can keep releasing features every day, but adoption is the constraint, which is part of why AI has not yet delivered ROI. He argues that adoption needs to be designed for rather than assumed.
“one of the old stats was like most people use about 25% of all the software tools that they have”
He says that shipping more features adds cost, as products get bloated and features may not work together well or be considered enough. He says that for Linear the aim has been the right features built in a streamlined way rather than the most features, which means the product is simpler and users spend less time setting things up.
“if AI makes building things faster, it doesn't mean you should be like, Shipping more stuff necessarily”
On 20VC, Alex Mashrabov said all eight creative directors Higgsfield interviewed named camera control, which became the product that took it from about $1M to $20M ARR in three months.
Grosser said no one has invented a test for product-market fit, so the question of product versus go-to-market keeps coming up. At Vercel she picked 10 large existing enterprise customers that used Vercel but not yet in front of their main site, and had the product manager and lead engineer hear all 10 in a row to build pattern recognition. Afterwards everyone agreed there were product gaps to close before the product would be much more easily sellable. She noted that some enterprises still buy before full fit, because a visionary buyer will always work through the pain.
“So we're like, great, let's go get 10 of those. Let's get the product manager, lead engineer, and let's run them through conversations with all 10 in a row, so they can get pattern recognition.”
After the pivot, Higgsfield asked eight creative directors what was missing from AI video, and every one of them said camera control, which they called essential to storytelling. The product launched on March 31 and Mashrabov describes product-market fit as immediate. He says VFX and camera control took Higgsfield from roughly $1M to $20M ARR in about the first three months.
“We spoke to eight creative directors about their experience with AI and what's simply missing. Everyone told us that camera control does not exist in AI.”
He said he told his board in July that he would hold 30 CFO conversations in three months, using a road show, dinners and Q and A sessions with guests such as Pablo Dominguez and Kyle Norton. He said he would record the conversations with AI to find patterns afterwards. He said changing buyer behaviour made listening more important than before.
“I told the board in July's board meeting, I am going to have 30 CFO conversations over the next three months.”
He calls mass-market consumer product-market fit extraordinarily difficult, especially when many talented, well-resourced incumbents already serve the market. Robinhood broke through by pairing its view of how people would use mobile with systematic processes for understanding customer needs and flowing them into what it built.
“through pretty systematic ways of talking to customers and understanding what their needs are, and, you know, flowing those into the product development.”
Starting around August 2009, for four months, Kerrest talked to IT professionals about their problems and whether they would use or buy a solution, and showed them drawings, documents and pictures. He tracked each name and contact date in green, yellow or red, and says he missed his number in some months. He cites Peter Thiel's Zero to One and Steve Blank's Four Steps to the Epiphany as the basis for the approach.
“I had a monthly plan where I had to go talk to 18 net new IT professionals and just talk to them and say, Hey, what kind of problems do you have? Do you have this problem? Would you use this? Would you buy this?”
Mike said LinkedIn used a process taught by Frank Robinson in which a product manager, an engineer and a sales or marketing person travelled in waves of five or six prospects and changed mock-ups back at the hotel for the next day. The aim was to exit with 10 or 15 prospects contractually committing to buy at a stated price if the product was built as described. He said the team spoke to 30 prospective clients in total.
“You go off and you speak to 30 prospective clients.”
Ranked by how many independent voices make each point and how specific their evidence is. Co-hosts of a show count as one voice, and a point needs at least two shows to appear here.
said Ann Davis (Revenue Builders), Chuck Ganapathi ([Un]Churned), Diane Wu ([Un]Churned), Brady Bluhm ([Un]Churned)
Ann says Crunchbase launched its MCP server about a month before the recording and that the value lies in bringing data sources together through an AI interface that sits inside the seller's workflow. She says the choice of front-end AI tool, such as ChatGPT, Gemini or Grok, does not matter.
“it's about how you seamlessly bring together the data sources through the AI interface to deliver real value.”
Chuck Ganapathi says Gainsight launched its MCP interface about a month before Pulse, and all its products now have one. He says this lets customers work in their preferred agentic tool and use Gainsight without ever having to log into it, and that the Gainsight CLI lets admins configure the product through coding agents.
“you can use Gainsight without ever having to log into Gainsight”
Diane Wu says that in the old world CSMs used multiple tabs and tools to get the context they need, and she expects this to go away. She describes the LLM as the new workspace, with the interface being natural language rather than visual, at least today. She adds that this might change very quickly in the future.
“I'm a believer of the traditional application UI is going to eventually fade and go away and LLMS will become the new workspace”
Brady Bluhm says he had a product manager's existential moment after playing with Staircase's MCP and asking why anyone would log into his product in a year and a half or two years. He expects that change because LLMs themselves are improving faster than other software. He still builds for the UI today because not everyone works in LLMs all the time.
“in two years from now, I think that's going to change because the LLMS themselves evolved so much faster than any other product will too”
said Mark Vovsi ([Un]Churned), Adam Liska (Topline), Steve Cox (Topline)
He said conversational access through Claude, MCP or similar suits quick questions, but he still sees a need for a platform where people can see information in one view. He said a summary from a chat tool can run to three pages, while a CSP organises the information and shows where to focus at the start of the day. He said he does not think the headless model has arrived yet.
“I think the future is a combination.”
Adam agrees most software will support headless access through APIs and MCP. He says people don't just connect a chat model to all their systems and push it to act; they need to be served the context and data required to figure out the next best step. He says UX and UI remain important and that the interface should differ for a sales manager, a rep and a CRO. He describes this as a more flexible, adaptable interface that can be rebuilt around each person's goal.
“But really, I think just saying Headless will solve everything, I think it's quite a, you know, it's an easy way to get out.”
He says the underlying logic and context layers remain the same, and that the company can build a skin suited to each segment. He says headless access lets customers consume the intelligence however they want, and that the company's MCP server is live and already gaining traction.
“I think underlying the logic layer and the context layer remains the same and then I think you're on top of that you have the opportunity to build a skin right that's suitable for you know SMB customers mid-market.”
Those predicting UI decline focus on fast-improving LLMs for quick queries, while defenders focus on users who need organised, role-specific views to decide daily priorities.
said Grant Clarke ([Un]Churned), Rick Smolen (Topline), Cassie Young (Topline), Diego Ballona ([Un]Churned)
Grant Clarke says Atlas records what reps do outside the agent's handoffs and classifies those actions into buckets. Forward deployed engineers, product managers and business analysts sit 'in the boat' with renewal managers to monitor this. When a category of escalation recurs multiple times, it becomes a product requirement. The FDE team develops it in maybe two weeks, tests it for a few weeks, and puts it back into Atlas.
“The FDE team can sprint, develop that in maybe two weeks, run a few weeks of testing, and then it's right back into Atlas to handle the next customer issue that comes up.”
Rick says a core offering may handle around 90% of what customers need, and that the remaining edge cases can be handled by building applications that complement the core product. He says this is more possible now, but that it adds complexity because forward-deployed engineering teams must work with core engineering. He says each organization should decide which things it will not build into the core product but will still say yes to.
“what are the things that we're not going to build into our core offering? But we should be able to say yes to”
Cassie says she is seeing a notable organizational change in portfolio companies: implementation, previously in the chief customer officer or post-sales lane, is moving into product. Companies view getting customers up and running as a product opportunity, and the best organizations productize what they learn from forward-deployed work immediately.
“implementation in many places is being moved out of the chief customer officer lane or the post-sales lane and into product”
A host says many AI companies have forward-deployed engineers but no plumbing connecting them to product, so the work is just implementation. He says the true FDE model is one where that work feeds back into what gets built, which is what made the Palantir model special. Cassie agrees, saying Primary has discussed the Palantir model internally and that Palantir built product off it rather than just selling services.
“The whole FTE concept was it was going to help Palantir figure out what to build and that feedback loop is what was so special about the whole model”
He said FDE works on early prototypes with specific customers. Once a product becomes a big part of the platform's infrastructure and needs more investment, a core product team takes ownership and productizes it for all customers.
“But at some point, it makes sense for a core product team to own the product because it becomes a big part of like our infrastructure of our platform.”
A customer had a large share of its support volume on Slack, which Fin did not support at the time. Engineers sat with the customer's head of support, and an engineer estimated a version could be built in two to three weeks. Intercom built it for that customer despite it not being on the roadmap, and because demand was validated, it was productized into a general availability product in about three months.
“we actually decided to invest in that because we validated demand.”
Diego said this is different from aggregated research or metrics, which he called very valuable but slower to act on. He said the FDE motion has led Intercom to build a better product for all customers and better internal tooling for its go-to-market team.
“when you get engineers sitting with customers directly, they feel the pain first hand.”
said Rebecca Nerad ([Un]Churned), Snehal Nimje (Topline), Steve Cox (Topline), Frederic Kerrest (The Science of Scaling)
Nerad described a CS director at another company who builds customized dashboards from product data for each customer, and said customers love it. Her concern is how that scales. She said ideas CSMs develop with customers must be fed back into the product, or the industry ends up where software was 25 years ago, with customizations it can't maintain.
“Else we're going to end up in the place where software was when I got into the industry 25 years ago with all the customizations that we aren't able to maintain too”
Snehal said the team would try to please any customer and ship requested features quickly, and each new request led to more requests. The result was many features and settings that made the product complicated from the product side and from the marketing side, which the team saw as part of the problem before narrowing focus.
“So it became more chaotic from the product side.”
He says not everybody is a good architect, and that many mid-market buyers do not want to stitch point solutions together. He argues the answer is to keep the product simple rather than rely on forward-deployed engineers and custom builds.
“just because you can build it doesn't mean you should”
Kerrest says reps trying to hit a number will insist a feature is needed to land a big customer and that everyone wants it, when it turns out to be a corner case only that customer needs, and possibly not even required to close the deal. Founders who aren't talking directly to customers lose the information needed to judge this.
“if we build this feature We're gonna get this big customer and everyone else wants it turns out no one else wants it”
Custom work pays off for enterprise vendors with a disciplined path back into the core product, but becomes bloat for mid-market sellers or teams that never productize it.
What one named guest described doing or seeing. Each is a single account, not a point several operators agree on.
“Your new business cost of sale contribution is 30 % of the total deal size. That's insane. How do you fix that without pissing off the sales team? That's a big question.”
“we can randomize both the orchestrator and the harness for the agent so we can collect all of those interaction effects that make it challenging.”
“And of course, the name your own price for an airline in Europe is a really bad idea.”
“you can shoot more end to end from the very beginning at scale, and I think doing so puts you on this path of rapid self-improvement.”
“We know that for an implementation for Vantagepoint for X number of employees, it should take around eight months.”
“in the investment bank when you trade on the interbank market, the minimum ticket size is like tens of millions”
Grosser said no one has invented a test for product-market fit, so the question of product versus go-to-market keeps coming up. At Vercel she picked 10 large existing enterprise customers that used Vercel but not yet in front of their main site, and had the product manager and lead engineer hear all 10 in a row to build pattern recognition. Afterwards everyone agreed there were product gaps to close before the product would be much more easily sellable. She noted that some enterprises still buy before full fit, because a visionary buyer will always work through the pain.
“So we're like, great, let's go get 10 of those. Let's get the product manager, lead engineer, and let's run them through conversations with all 10 in a row, so they can get pattern recognition.”
Azam said engineering organizations easily slip into a pattern where engineers do not talk to customers because product is the first layer. He said the best engineers want direct contact, and a direct conversation with a practitioner from time to time beats feedback relayed through three abstracted organizations. A host added that such engineers then change their language with sellers, helping them stand in the user's moment of pain.
“there's nothing like getting an engineer to speak to a practitioner who's using their product directly”
Roberge's example: tell the prospect the product launches in two months at $50,000 a year, and that you want five design partners to sign a letter of intent at a 90% discount, $5,000. Asking them to commit money, drawing on Steve Blank's idea that the hardest thing is separating someone from their wallet, reveals whether a stated 'yes, I'd use this' is real. He applies this before the product exists.
“But by pushing them to separate them from their wallet gives us the real feedback. It's a differentiation between a false positive and a true positive.”
Josh Schachter describes design-thinking user research from his time at BCG. Asking 'how often do you do X?' got thin answers like 'every other month'. Following up with 'tell me the last time that happened, give me that situation' put people in a concrete scene. He links this to London's hypothetical scenes, saying both let the person picture a situation.
“okay, great, well, tell me the last time that that happened. Like, give me that situation.”
Grant Clarke says Atlas records what reps do outside the agent's handoffs and classifies those actions into buckets. Forward deployed engineers, product managers and business analysts sit 'in the boat' with renewal managers to monitor this. When a category of escalation recurs multiple times, it becomes a product requirement. The FDE team develops it in maybe two weeks, tests it for a few weeks, and puts it back into Atlas.
“The FDE team can sprint, develop that in maybe two weeks, run a few weeks of testing, and then it's right back into Atlas to handle the next customer issue that comes up.”
He said FDE works on early prototypes with specific customers. Once a product becomes a big part of the platform's infrastructure and needs more investment, a core product team takes ownership and productizes it for all customers.
“But at some point, it makes sense for a core product team to own the product because it becomes a big part of like our infrastructure of our platform.”
A customer had a large share of its support volume on Slack, which Fin did not support at the time. Engineers sat with the customer's head of support, and an engineer estimated a version could be built in two to three weeks. Intercom built it for that customer despite it not being on the roadmap, and because demand was validated, it was productized into a general availability product in about three months.
“we actually decided to invest in that because we validated demand.”
Jo Massie said the team asks how much time and budget a project deserves, such as one, two or four weeks. She said anything taking more than four weeks to implement is quite expensive, so the team needs to see results from it. She said that if an idea does not work, the team should not invest too much in it, and that small pilots are the way to start.
“How much time are we willing to give this one week, 2 weeks, 4 weeks?”
Bryan said each idea is framed around a customer problem, given a set number of weeks to prove out, and measured against a confidence level. He said they give a little latitude but then move on to the next idea on the list, so they keep moving forward. He said that once an idea reaches a high confidence level, giving 80% as an example, it moves into the product and engineering roadmap.
“and you're going to be able to measure it with a confidence level”
The census tool went from a bought wrapper tool, to an internal build, to a product-team build into the Justworks product, going live after about a year. Internal productivity tools generally stay with RevOps and go-to-market engineering, but some belong in the product, especially for self-service. Lauren Hughes said there was push and pull with product in reaching this approach.
“We've kind of taken the stance that revenue operations will stand up something in the short term to increase immediate productivity on the floor. And we'll buy air cover for our product team to develop that the right way over time”
Usha says HiveBright has robust APIs but had no pre-built integrations, so she proposed getting a middleware platform and a contractor to build a few integrations, noting that a connector such as Salesforce or HubSpot does not take much. They built a couple, pitched them and saw traction, and over 30% of customers now buy them. The product team has since taken them over.
“now over 30 % of our customers buy pre -built integrations from us.”
Basic product analytics asks whether customers are using the product at all. More sophisticated tracking asks what use case a customer should be using the product for and whether they use the features associated with it. If they do not, Christine says the team can advise them on how to use the product to reach their outcomes. She describes product analytics as her first 2026 priority, with cross-functional work underway.
“The more complex or sophisticated ones are around what's the use case that a customer should be using our product for, and are they using the features associated with that particular use case?”
Rob said that for downmarket customers, trying to adopt AI features early did not correlate well with a good outcome. He said getting some fundamental data structures right as part of onboarding was important before applying the AI features. Ironclad now uses these findings to decide where to intervene.
“for for AI features within our product, if they tried to adopt them too soon, it actually didn't correlate well to a good outcome.”
He says he tells the R&D team to reward building new things as much as throwing code out. His reasoning is that without deleting code, the stack slowly turns into a legacy software stack. He describes this as a practice he asks of his own team at Glean.
“reward, building new stuff as much at the same level as throwing some stuff out of your code base. Because if you're not throwing things out, it means you're sort of slowly and slowly converting into a legacy software stack.”
A former solutions engineer designed the system, and CSMs use it constantly. An intake agent listens to Gong calls and monitors a Slack channel where CSMs post requests. Another agent restates the request in natural language, another checks Salesforce for whether the requester is a customer, and another logs it in a database. That gives a top-10 feature request list across customers, and requests are referenced in customers' AI transformation plans. The next step Christina Parra described as planned is an automatic alert to the customer when a request ships, to build trust.
“There's another one that checks Salesforce to see, is this a customer, is this not a customer?”
He said a CSM should grab feedback from a customer, validate that it is important, combine it across the entire portfolio, and have it delivered to the product team. This lets the product team prioritize what unlocks the business. He presented this as the mechanism for handling incomplete features in the health model.
“let me grab the feedback from my customer, validate that it actually is important, combine it across the entire portfolio and have that delivered to our product team so you can prioritize what unlocks the business.”
He describes a pharmacy's conversational bot with a two second response target, which he calls way too slow for conversation, and says conversational use needs roughly 500 milliseconds or less. He says Cerebras powers instant AI for conversational, agentic coding and deep search use cases. He describes asking a prospect what changes if responses arrive in 200 milliseconds against a 500 millisecond target.
“You need to be like 500 milliseconds or less, right?”
Amanda Kahlow says the live demo product sits inside a customer's product and carries a large amount of context, which makes latency hard to keep down. Her company uses filler words and holding phrases to make the response feel instant while the system works through a hard reasoning problem. She says they measure accuracy, latency, speed and cost in multi-agent orchestration.
“Now there's a lot of things you can do to fake it. Like, hey, let me pull that up for you.”
said Ann Davis (Revenue Builders), Chuck Ganapathi ([Un]Churned), Diane Wu ([Un]Churned), Brady Bluhm ([Un]Churned)
Ann says Crunchbase launched its MCP server about a month before the recording and that the value lies in bringing data sources together through an AI interface that sits inside the seller's workflow. She says the choice of front-end AI tool, such as ChatGPT, Gemini or Grok, does not matter.
“it's about how you seamlessly bring together the data sources through the AI interface to deliver real value.”
Chuck Ganapathi says Gainsight launched its MCP interface about a month before Pulse, and all its products now have one. He says this lets customers work in their preferred agentic tool and use Gainsight without ever having to log into it, and that the Gainsight CLI lets admins configure the product through coding agents.
“you can use Gainsight without ever having to log into Gainsight”
Diane Wu says that in the old world CSMs used multiple tabs and tools to get the context they need, and she expects this to go away. She describes the LLM as the new workspace, with the interface being natural language rather than visual, at least today. She adds that this might change very quickly in the future.
“I'm a believer of the traditional application UI is going to eventually fade and go away and LLMS will become the new workspace”
Brady Bluhm says he had a product manager's existential moment after playing with Staircase's MCP and asking why anyone would log into his product in a year and a half or two years. He expects that change because LLMs themselves are improving faster than other software. He still builds for the UI today because not everyone works in LLMs all the time.
“in two years from now, I think that's going to change because the LLMS themselves evolved so much faster than any other product will too”
said Grant Clarke ([Un]Churned), Rick Smolen (Topline), Cassie Young (Topline), Diego Ballona ([Un]Churned)
Grant Clarke says Atlas records what reps do outside the agent's handoffs and classifies those actions into buckets. Forward deployed engineers, product managers and business analysts sit 'in the boat' with renewal managers to monitor this. When a category of escalation recurs multiple times, it becomes a product requirement. The FDE team develops it in maybe two weeks, tests it for a few weeks, and puts it back into Atlas.
“The FDE team can sprint, develop that in maybe two weeks, run a few weeks of testing, and then it's right back into Atlas to handle the next customer issue that comes up.”
Rick says a core offering may handle around 90% of what customers need, and that the remaining edge cases can be handled by building applications that complement the core product. He says this is more possible now, but that it adds complexity because forward-deployed engineering teams must work with core engineering. He says each organization should decide which things it will not build into the core product but will still say yes to.
“what are the things that we're not going to build into our core offering? But we should be able to say yes to”
Cassie says she is seeing a notable organizational change in portfolio companies: implementation, previously in the chief customer officer or post-sales lane, is moving into product. Companies view getting customers up and running as a product opportunity, and the best organizations productize what they learn from forward-deployed work immediately.
“implementation in many places is being moved out of the chief customer officer lane or the post-sales lane and into product”
A host says many AI companies have forward-deployed engineers but no plumbing connecting them to product, so the work is just implementation. He says the true FDE model is one where that work feeds back into what gets built, which is what made the Palantir model special. Cassie agrees, saying Primary has discussed the Palantir model internally and that Palantir built product off it rather than just selling services.
“The whole FTE concept was it was going to help Palantir figure out what to build and that feedback loop is what was so special about the whole model”
He said FDE works on early prototypes with specific customers. Once a product becomes a big part of the platform's infrastructure and needs more investment, a core product team takes ownership and productizes it for all customers.
“But at some point, it makes sense for a core product team to own the product because it becomes a big part of like our infrastructure of our platform.”
A customer had a large share of its support volume on Slack, which Fin did not support at the time. Engineers sat with the customer's head of support, and an engineer estimated a version could be built in two to three weeks. Intercom built it for that customer despite it not being on the roadmap, and because demand was validated, it was productized into a general availability product in about three months.
“we actually decided to invest in that because we validated demand.”
Diego said this is different from aggregated research or metrics, which he called very valuable but slower to act on. He said the FDE motion has led Intercom to build a better product for all customers and better internal tooling for its go-to-market team.
“when you get engineers sitting with customers directly, they feel the pain first hand.”
said Dan Lee (The Science of Scaling), Keith Peiris (Topline), Steve Cox (Topline), Karri Saarinen (Grit)
Dan Lee said it is easy to build a ton of features, so the hard part is knowing what is important. Nooks lives by a do more with less value, and he described its approach as rigorously evaluating options against each other and setting clear north-star metrics and leading indicators.
“we can do anything but not everything.”
A host had observed that GTM platforms bolt features onto a left-hand menu without connected workflows. Keith agreed that everyone is racing to the roughly 100 features required to run a go-to-market team. He said making them performant and coherent together is really hard, and many players produce apps that in theory prospect, forecast and hold a CRM data model but aren't usable by a rep.
“create this Frankenstein app. You know, with a hundred features that in theory could prospect, in theory could forecast, in theory has a CRM data model, but is not a thing that a rep can use.”
He says companies can keep releasing features every day, but adoption is the constraint, which is part of why AI has not yet delivered ROI. He argues that adoption needs to be designed for rather than assumed.
“one of the old stats was like most people use about 25% of all the software tools that they have”
He says that shipping more features adds cost, as products get bloated and features may not work together well or be considered enough. He says that for Linear the aim has been the right features built in a streamlined way rather than the most features, which means the product is simpler and users spend less time setting things up.
“if AI makes building things faster, it doesn't mean you should be like, Shipping more stuff necessarily”
said Jeanne DeWitt Grosser (Grit), Alex Mashrabov (The Twenty Minute VC), AJ Bruno (Topline), Baiju Bhatt (Grit), Frederic Kerrest (The Science of Scaling)
Grosser said no one has invented a test for product-market fit, so the question of product versus go-to-market keeps coming up. At Vercel she picked 10 large existing enterprise customers that used Vercel but not yet in front of their main site, and had the product manager and lead engineer hear all 10 in a row to build pattern recognition. Afterwards everyone agreed there were product gaps to close before the product would be much more easily sellable. She noted that some enterprises still buy before full fit, because a visionary buyer will always work through the pain.
“So we're like, great, let's go get 10 of those. Let's get the product manager, lead engineer, and let's run them through conversations with all 10 in a row, so they can get pattern recognition.”
After the pivot, Higgsfield asked eight creative directors what was missing from AI video, and every one of them said camera control, which they called essential to storytelling. The product launched on March 31 and Mashrabov describes product-market fit as immediate. He says VFX and camera control took Higgsfield from roughly $1M to $20M ARR in about the first three months.
“We spoke to eight creative directors about their experience with AI and what's simply missing. Everyone told us that camera control does not exist in AI.”
He said he told his board in July that he would hold 30 CFO conversations in three months, using a road show, dinners and Q and A sessions with guests such as Pablo Dominguez and Kyle Norton. He said he would record the conversations with AI to find patterns afterwards. He said changing buyer behaviour made listening more important than before.
“I told the board in July's board meeting, I am going to have 30 CFO conversations over the next three months.”
He calls mass-market consumer product-market fit extraordinarily difficult, especially when many talented, well-resourced incumbents already serve the market. Robinhood broke through by pairing its view of how people would use mobile with systematic processes for understanding customer needs and flowing them into what it built.
“through pretty systematic ways of talking to customers and understanding what their needs are, and, you know, flowing those into the product development.”
Starting around August 2009, for four months, Kerrest talked to IT professionals about their problems and whether they would use or buy a solution, and showed them drawings, documents and pictures. He tracked each name and contact date in green, yellow or red, and says he missed his number in some months. He cites Peter Thiel's Zero to One and Steve Blank's Four Steps to the Epiphany as the basis for the approach.
“I had a monthly plan where I had to go talk to 18 net new IT professionals and just talk to them and say, Hey, what kind of problems do you have? Do you have this problem? Would you use this? Would you buy this?”
Mike said LinkedIn used a process taught by Frank Robinson in which a product manager, an engineer and a sales or marketing person travelled in waves of five or six prospects and changed mock-ups back at the hotel for the next day. The aim was to exit with 10 or 15 prospects contractually committing to buy at a stated price if the product was built as described. He said the team spoke to 30 prospective clients in total.
“You go off and you speak to 30 prospective clients.”
said Rebecca Nerad ([Un]Churned), Snehal Nimje (Topline), Steve Cox (Topline), Frederic Kerrest (The Science of Scaling)
Nerad described a CS director at another company who builds customized dashboards from product data for each customer, and said customers love it. Her concern is how that scales. She said ideas CSMs develop with customers must be fed back into the product, or the industry ends up where software was 25 years ago, with customizations it can't maintain.
“Else we're going to end up in the place where software was when I got into the industry 25 years ago with all the customizations that we aren't able to maintain too”
Snehal said the team would try to please any customer and ship requested features quickly, and each new request led to more requests. The result was many features and settings that made the product complicated from the product side and from the marketing side, which the team saw as part of the problem before narrowing focus.
“So it became more chaotic from the product side.”
He says not everybody is a good architect, and that many mid-market buyers do not want to stitch point solutions together. He argues the answer is to keep the product simple rather than rely on forward-deployed engineers and custom builds.
“just because you can build it doesn't mean you should”
Kerrest says reps trying to hit a number will insist a feature is needed to land a big customer and that everyone wants it, when it turns out to be a corner case only that customer needs, and possibly not even required to close the deal. Founders who aren't talking directly to customers lose the information needed to judge this.
“if we build this feature We're gonna get this big customer and everyone else wants it turns out no one else wants it”
said Dan Lee (The Science of Scaling), Lou Shipley (Revenue Builders), Sahir Azam (Revenue Builders)
Dan Lee said that when Nooks went all in on sales, he spent 60 to 70 percent of his week observing customers using the product, which he described as the majority of his time. The team did this by spending time on the sales floor with the sales teams using Nooks.
“Um definitely the majority like 60 70%”
Lou was selling in Japan when a customer placed a large order with a couple of requested features and offered to buy 10 more, which would have been the company's biggest order. He could not work out what the customer wanted to do with the product, so he took his top rep, Yoshiro Takahashi, a former car salesman, to a karaoke bar. They drank beer and watched karaoke with a legal pad, and he worked out which features to build, which led to the biggest order in the company's history.
“I went with a legal pad and a pen. And we drank beer. And watched karaoke and I finally figured out what features this customer wanted”
Lou says that across his six startups, customers did things with the products that he had no idea they would do. He says that when you sit down and interview customers you learn what they are using the product for. He presents this intellectual curiosity as the key to understanding use cases.
“customers did things with the products. I had no idea they were going to do it.”
Azam said engineering organizations easily slip into a pattern where engineers do not talk to customers because product is the first layer. He said the best engineers want direct contact, and a direct conversation with a practitioner from time to time beats feedback relayed through three abstracted organizations. A host added that such engineers then change their language with sellers, helping them stand in the user's moment of pain.
“there's nothing like getting an engineer to speak to a practitioner who's using their product directly”
said Mark Vovsi ([Un]Churned), Adam Liska (Topline), Steve Cox (Topline)
He said conversational access through Claude, MCP or similar suits quick questions, but he still sees a need for a platform where people can see information in one view. He said a summary from a chat tool can run to three pages, while a CSP organises the information and shows where to focus at the start of the day. He said he does not think the headless model has arrived yet.
“I think the future is a combination.”
Adam agrees most software will support headless access through APIs and MCP. He says people don't just connect a chat model to all their systems and push it to act; they need to be served the context and data required to figure out the next best step. He says UX and UI remain important and that the interface should differ for a sales manager, a rep and a CRO. He describes this as a more flexible, adaptable interface that can be rebuilt around each person's goal.
“But really, I think just saying Headless will solve everything, I think it's quite a, you know, it's an easy way to get out.”
He says the underlying logic and context layers remain the same, and that the company can build a skin suited to each segment. He says headless access lets customers consume the intelligence however they want, and that the company's MCP server is live and already gaining traction.
“I think underlying the logic layer and the context layer remains the same and then I think you're on top of that you have the opportunity to build a skin right that's suitable for you know SMB customers mid-market.”
said Christina Parra ([Un]Churned), Christopher O'Donnell (The Science of Scaling), Abbas Haider Ali ([Un]Churned)
A former solutions engineer designed the system, and CSMs use it constantly. An intake agent listens to Gong calls and monitors a Slack channel where CSMs post requests. Another agent restates the request in natural language, another checks Salesforce for whether the requester is a customer, and another logs it in a database. That gives a top-10 feature request list across customers, and requests are referenced in customers' AI transformation plans. The next step Christina Parra described as planned is an automatic alert to the customer when a request ships, to build trust.
“There's another one that checks Salesforce to see, is this a customer, is this not a customer?”
Asked for one tactic for young engineering founders building in a vacuum, Christopher (acknowledging he is promoting Day AI) suggests recording meetings and connecting Slack into a customer memory layer. From there feedback can flow automatically to GitHub, or be worked through Claude Code. He says the expectation should be past tickets: automated action, rolled-up strategic insights, and answers to questions like who a feature would affect or what the biggest gaps are for a new persona.
“I think we're definitely past the kind of ticket concept, you know, level and you should be able to have some sort of automated action off of that”
He said a CSM should grab feedback from a customer, validate that it is important, combine it across the entire portfolio, and have it delivered to the product team. This lets the product team prioritize what unlocks the business. He presented this as the mechanism for handling incomplete features in the health model.
“let me grab the feedback from my customer, validate that it actually is important, combine it across the entire portfolio and have that delivered to our product team so you can prioritize what unlocks the business.”
Paul says engineers pushed back when sales asked for a feature for a single deal. He says the request was more persuasive when it showed a segment of developers who could not use GitHub without the change, such as a large group in government or finance.
“if you came and said, I need to get this feature for one deal, they'd shoot you. But if you came back and you said, there's 500,000 developers sitting inside the government sector”
said Parag Agarwal (Grit), Arvind Jain (Grit), Wade Foster (Topline)
At Twitter he says they had to rebuild entire systems roughly every year because the systems weren't built for the next order of magnitude. He has carried the same lessons to Parallel, which he expects to become meaningfully larger in scale than Twitter: design pragmatically for the current scale and know when to rebuild.
“if you build for like four years out, you don't ship anything. And so you have to constantly keep evolving for scale.”
He says he tells the R&D team to reward building new things as much as throwing code out. His reasoning is that without deleting code, the stack slowly turns into a legacy software stack. He describes this as a practice he asks of his own team at Glean.
“reward, building new stuff as much at the same level as throwing some stuff out of your code base. Because if you're not throwing things out, it means you're sort of slowly and slowly converting into a legacy software stack.”
He says the technology stack is evolving faster than anything he has seen, so he assumes a product built a year earlier can be done better now. He frames not finding a better approach as a failure of imagination rather than a sign the work is finished. He presents this as his own default mindset.
“My mindset by default is that if you build something last year, that it's got to be obsolete. There has to be a new way to do that thing better today. If not, then it's just lack of imagination.”
Wade said that if you talk to engineers at a company, they probably have something like Cursor or Claude Code running constantly. Teams build prototypes to test ideas quickly and discard many of them. He said the cost of throwing away code no longer feels like a big deal.
“And the cost of throwing away code, it just doesn't feel like that big of a deal.”
said Bryan Murphy (Topline), Jo Massie ([Un]Churned)
Bryan said each idea is framed around a customer problem, given a set number of weeks to prove out, and measured against a confidence level. He said they give a little latitude but then move on to the next idea on the list, so they keep moving forward. He said that once an idea reaches a high confidence level, giving 80% as an example, it moves into the product and engineering roadmap.
“and you're going to be able to measure it with a confidence level”
Jo Massie said the team asks how much time and budget a project deserves, such as one, two or four weeks. She said anything taking more than four weeks to implement is quite expensive, so the team needs to see results from it. She said that if an idea does not work, the team should not invest too much in it, and that small pilots are the way to start.
“How much time are we willing to give this one week, 2 weeks, 4 weeks?”
said Christopher O'Donnell (The Science of Scaling), Liz Christo (Topline), Jeremey Donovan (The Revenue Leadership Podcast)
Christopher says a single-user CRM needs 45 minutes, not a weekend; once permissions over who sees whose leads come in, people buy HubSpot or Salesforce. Customer memory is a superset of CRM and gets 'really intense really quickly': it has to populate almost instantly and be built for the AI. Teams building it themselves reach for vector databases and property graphs rather than relational databases, and he thinks they will ultimately decide to use a vendor.
“if you're just trying to build yourself a CRM, you don't need a weekend. You need 45 minutes. The minute you go multiplayer.”
Liz says the costs of versions two through ten, maintenance, edge cases, security and compliance are buried behind the scenes because building with AI is still sexy and fun. She thinks the current build-heavy phase will come back around once those costs surface.
“there's like a huge amount of cost buried behind the scenes that we're not really talking about today because it's still like sexy and fun”
He used customer health as an example: the health score is a traditional predictive ML problem, drafting messages for CSMs is a generative task, and turning a flagged account into a workflow is agentic. He said this heterogeneity is one reason companies may skew toward platforms and point solutions rather than building.
“Like that stuff starts to get complicated.”
said Jason Forget (Revenue Builders), Asad Zaman (Topline), Usha Iyer (The Revenue Leadership Podcast)
One host said early-stage sellers are part product manager because the product will not yet do everything it claims, so they must sell around gaps while giving honest customer feedback to development. Jason agreed and said this becomes less necessary once the company has product marketing and product management teams. He called the ability gold for the business.
“in the early days, the salespeople are part salespeople, part product manager.”
Asad Zaman describes a founder with about $30M NR who merged engineering, product and design into one unit working closely with go-to-market. The go-to-market team listens to the market and feeds the product team, which can build and ship quickly, so the company stays one step ahead of competitors who can copy features within days.
“there are no product advantages that are longlasting anymore”
Usha says product teams have to rely on second-hand information when they are not close to customers. She recommends aligning product work to each step of the customer journey so that GTM teams are involved in experiments from the start, which she says would help product teams work faster.
“Their experiments can bring go -to -market people into their experiments from the beginning.”
Daniel, one of the first two reps Greg hired, went from his calls straight to the engineering office with what he was hearing, and pushed for the product to connect video endpoints with browsers and Skype. Greg says the engineering team built the first version, called the room connector, within about three months, and the company started winning against BlueJeans, which had been connecting endpoints.
“Within like three months we had like first version of this we called it the room connector and all of a sudden we started knocking out knocking out BlueJeans”
After learning that customers loved the analytics part of the customer data platform so much they wanted to buy it on its own, Klaviyo set up a twice-a-week standup for the customer growth team. Within four weeks it created an analytics-only SKU, because product marketing, R&D and the other needed functions were on the standups listening to calls and hearing feedback.
“But within four weeks, we were able to create a new SKU for analytics only because we had product marketing, we had R&D, we had everyone that was needed to get this done on those stand ups, listening to these calls, hearing the feedback.”
Jeff Perry said the executive team took the LLC signal from the market and asked the product teams whether something should be developed differently for LLCs. They found features such as K1 distributions made it a different product set, so they dedicated a product team to building it. He called this a way to use the relationship between sales feedback and a direct line to product.
“From there, it became, OK, take the signal you're getting and involve the product teams and say, is there something we should develop differently because of the needs of an LLC?”
said Alex Mashrabov (The Twenty Minute VC), Aman Narang (Grit)
Mashrabov says Higgsfield spent more than a year searching for a product that worked, and he takes responsibility for optimizing for hype, narrative and attention instead of building a good product. With slightly less than $5M left and feeling they had 'one attempt left', the team committed to product and PLG on the belief that the best product would win. That led to the launch that took off.
“I was so much optimizing for what's hype today, what's the right narrative, how we can hijack the attention, all these things, really. Everything instead of building a good product.”
After early customers arrived, investor Steve Papa pushed the team to put the pedal down, but Aman said the software did not work offline, the hardware was not ready for deployment, and support was not in place. He recalled that 100% of the first hardware iteration came back within roughly 80 days. Steve Papa and co-founder Steve Fredette told the team to slow down and fix these foundations.
“We thought we were ready to scale, but because of the complexity of the hardware, the software, the support, we just really were not ready.”
Mark describes a typical plan: build the product from January, train all hundred salespeople and the support and success teams, then launch at a July customer conference. He says the product then fails because it was not right yet, and argues teams should iterate with agile development and customer-centric design until product-market fit before scaling.
“We train all hundred salespeople. The customer success team and customer support team all trained on the new product. And we launch it and the product sucks.”
said Anastasios Angelopoulos (Grit), AJ Meyer (Topline)
Angelopoulos says the biggest historical criticism was that people vote for what they like or what confirms their biases, not for verifiable, factual or high-quality code. Arena has largely moved away from battles as the main signal. Task completion, which can be measured from the data without users stating it, is now number one, and he says it closes the gap between stated and revealed preferences. Steerability, hallucination rates and factuality are also ranked.
“we've moved largely away from battles as the main signal on arena for this reason, we incorporate human preference still. But the number one signal that we have now is task completion.”
Angelopoulos describes Arena as a model-agnostic ChatGPT or Claude Cowork, where users should never feel they have to give feedback. Explicit conversational feedback ('keep going', 'undo it') and implicit signals both feed the rankings. The implicit signals include query reformulation (as in search), downloads, whether files are actually used, and which PRs get merged. He argues this aligns user and platform incentives and produces the highest-quality model performance signal on the market.
“Every interaction is a piece of feedback. Download button. That's a piece of feedback.”
He says standard computer vision metrics such as intersection over union and average precision mean little to a customer. Pickle instead uses metrics such as boxes per hour dropped or conveyed, uptime and flow variability, and trains its models to optimize them, which encodes knowledge of the customer into the objective.
“What Pickle started doing was saying, what metrics does the customer care about?”
said Christopher O'Donnell (The Science of Scaling), Ghazi Masood ([Un]Churned)
Christopher says there is no need for an analyst to quantify the problem or a designer to mock up options. Someone on the phones and someone in product sit together over coffee, find something, verify its impact and ship it. In his office, four or five people may talk for three hours around couches over data and customer feedback, and by the end the software is written.
“You need somebody on the phone and somebody in product to sit together over a coffee and get a lead on something you could do, verify that it's going to have this impact and ship it 45 minutes later.”
Christopher says that in 2011-2012 he was PM, support expert, user researcher, go-to-market lead, front-end developer and designer on his products at HubSpot. By 2016 those were separate teams, and they kept splitting into roles like UX copy. Teams meant to be three people became seven, eight or nine, 'like a Quaker meeting', and lost effectiveness. Now PMs are doing sophisticated coding.
“And these small autonomous teams that we wanted to have all of a sudden were seven, eight, nine people.”
Ghazi says the leader had a vision, got input from his team on what they wanted, and built a prototype without waiting for engineers. The team then spent time refining and iterating on it before it was ready for wider use.
“had a prototype within hours”
said Keith Peiris (Topline), Chuck Ganapathi ([Un]Churned)
Keith admitted Lightfield doesn't yet have the best commission planning tool. His broader point is that customers are building CPQ in Lightfield themselves, using automations, conditional fields and forms, rather than buying a separate product.
“a lot of these features are now just automations and skills that run on your system of record in the new world.”
Chuck Ganapathi says the market frames the choice as either buying an inflexible piece of software that isn't tuned to your business or building it yourself, which is fun until you are stuck maintaining it. He says Gainsight lets customers build on its platform, which provides permissions, security, governance and data models, buy its prebuilt agents, or do both.
“we think that the right answer is that you should be able to build and buy”
said AJ Bruno (Topline), Garrett Marker ([Un]Churned), Brian McCarthy (Revenue Builders)
AJ said Gong's landing page now leads with 'Gong AI, ask anything' and pointed to the newly released Gong Bridge. He said this lets users find all their information across accounts quickly, rather than going to separate forecasting or call tools. He then asked Keith how Lightfield ties its own capabilities together.
“It's not about just the forecasting or the calls or whatever. It brings everything together.”
He said most tools focus on making an individual seller more effective, eliminating note-taking, or pushing data into the CRM. He argues that customer outcomes depend on the whole lifecycle, so the data needed to judge a buyer journey has to come from sales, CS and support together. This is his stated reason for building the narrative platform.
“I don't find that most tools are thinking about it. That way, I think they're thinking about it more for like an individual team”
Brian said Cursor bought Graphite and integrated its code review product, which he described as a team sport with multiple reviewers and therefore sticky. He said the next steps are the design, secure the code and implement the code areas, which would make Cursor an enduring software development lifecycle system.
“Code review is a team sport.”
Halligan describes Jobs explaining that MP3 players were too complicated for non-developers, so Apple made one simple device with a jukebox and bought songs one at a time. Halligan says this was one of many data points that led to HubSpot combining many separate marketing tools into one product.
“What if we could pull it all together and do one plus one plus one equals 10.”
said Chris Degnan (Revenue Builders), Ed Calnan (The Science of Scaling)
Localytics ran a 500-plus terabyte Vertica cluster on Amazon and needed data clustered in a way Snowflake did not support. Chris Degnan said the rebuild took the engineering team about seven months. The work let Snowflake turn on a recluster feature that made it faster and cheaper than open-source tools, and Nielsen then signed, which he called material.
“They had a 500 plus terabyte Vertica cluster on Amazon.”
Ed said the team had built a wizard that let individual users create their own sales collateral. A customer asked whether an administrator could generate materials for everything in batch and push them to sellers. Ed said the product did not need much change to do that.
“And we didn't have to change much for the product. That's the key.”
said Kyle Lacy (Topline), Kevin Mandia (Grit)
Kyle Lacy said that five years ago his main issue was getting products and engineering to ship fast enough. He said that is no longer the case because of AI, and that he now struggles to keep up with the volume of product releases, which makes product marketing more important in a product-led company.
“This is the first time in my career Where I'm having a hard time keeping up with the amount of stuff products and engineering is shipping”
Mandia says at Armadin engineering is ahead of marketing, which he contrasts with the usual pattern where marketing gets well ahead of the product. He says they plan to hold a demo every Friday and update the sales team on where the product is. He expects a much more rapid cycle between engineering and what is taken to market.
“For the first time in my career, engineering is ahead of marketing.”
said Jaleh Rezaei (Topline), Samantha Murray ([Un]Churned)
Jaleh said PLG requires prioritizing a better product because there is no one to show users how to navigate it. In her sales-led experience, teams kept shipping new features and relied on customer success to explain them. The point is that PLG pushes product quality earlier.
“it forces you to operate in a more prioritized way because you have to build a better, more complete product out of the box versus when we were building sales led.”
Jaleh said salespeople are unlikely to spend more than a few minutes on a new tool if it does not deliver value. Mutiny's internal mindset is that each user must earn the next 60 seconds, or they are gone. She presented this as the reason product-led growth forces a more complete product.
“you have 60 seconds.”
Samantha Murray says she spent almost seven years at Shopify, which had a large PLG motion, and that in-app optimization was the core of its experience team. She says that a PLG company makes no revenue unless it optimizes its first trial, so the sophistication she sees is far ahead of non-PLG companies.
“you don't make revenue unless you optimize your first 14-day trial or however long your trial is.”
said Christopher O'Donnell (The Science of Scaling), Karri Saarinen (Grit)
Alongside the rising emphasis on taste, he says design is converging back to a set of clean base defaults. The goal is doing the thing people expect most, not inventing something novel.
“It's not about doing something new. It's about doing the thing that people really expect the most.”
He says that for an application people use every day, changing the UI constantly is distracting because people are trying to do their jobs and the buttons keep moving. He suggests changing and testing new features is fine, but core flows and the main interface should not change every week, and that the software industry could use more patience about this.
“if you change the UI every day, people eventually will like hate it”
said Lauren Hughes (The Revenue Leadership Podcast), Usha Iyer (The Revenue Leadership Podcast)
The census tool went from a bought wrapper tool, to an internal build, to a product-team build into the Justworks product, going live after about a year. Internal productivity tools generally stay with RevOps and go-to-market engineering, but some belong in the product, especially for self-service. Lauren Hughes said there was push and pull with product in reaching this approach.
“We've kind of taken the stance that revenue operations will stand up something in the short term to increase immediate productivity on the floor. And we'll buy air cover for our product team to develop that the right way over time”
Usha says HiveBright has robust APIs but had no pre-built integrations, so she proposed getting a middleware platform and a contractor to build a few integrations, noting that a connector such as Salesforce or HubSpot does not take much. They built a couple, pitched them and saw traction, and over 30% of customers now buy them. The product team has since taken them over.
“now over 30 % of our customers buy pre -built integrations from us.”
said Justin Shriber (Topline), Steve Cox (Topline)
He said the operational insight was to meet people where they are, which means delivering insights in Slack and email, or on the call through the call-recording system. He said the era of proprietary apps is over for this use case.
“It's not about proprietary apps anymore. It's about Slack. It's about email.”
He says everyone now knows AI exists and will drive productivity, but many buyers do not know how to use it, so driving real adoption has become more important. He argues the workflow stays largely the same, so AI should be embedded into that workflow rather than added as more point solutions.
“I think the adoption and driving adoption has become more important than ever. I don't think throwing more tools is the answer.”
Actions written 10 Oct 2026 from the most useful of 198 recent insights and checked against them.
Sam describes Gold as an elite membership that sits on top of Pavilion's roughly 10,000 members. Candidates go through an interview process, CEOs and service providers are excluded, and fractionals must be in an between-role state rather than fractional for more than six months. Advisers can qualify after leaving an operating role such as at Databricks or OpenAI, but consultants generally do not.
“It is for operators, current operators from 50 million in ARR up to a billion and sometimes north of a billion.”
Braden describes the president's announcement of the Ag Passport. It records service projects, study abroad, involvement in 1,300 student organizations and civic activity. The aim is to show employers what that experience looks like at graduation and to set the expectation that students gain it while enrolled.
“announcing what we're calling the Ag Passport, which is basically documenting all of the education a student gets outside of the classroom.”
AJ Bruno says QuotaPath sits as connective tissue between CRM and HRIS, enriched by ERP data, and literally runs a customer's comp and delivers what the payout should be at quarter end. He argues services win on the extra value the customer didn't expect, such as flagging that BDRs are paid at twice the benchmark or that new-business cost of sale is 30% of total deal size. The hard follow-on question is how to fix such issues without upsetting the sales team.
“Your new business cost of sale contribution is 30 % of the total deal size. That's insane. How do you fix that without pissing off the sales team? That's a big question.”
AJ Bruno says QuotaPath treated plan building as a workflow of components rather than spreadsheet-style if-then logic. That gives visibility that differentiates it, but as companies adopt things like cumulative quotas, draws and multi-field earnings rules, QuotaPath must continuously build product to match, whereas a spreadsheet could simply add more if-then rules. Over eight years it matured to handle complex needs such as ASC 606 revenue recognition.
“So there was visibility at the dual level, which is a big differentiator for us, but also causes a lot of challenge when we have to actually continuously build to the market where a spreadsheet could just be, if then, if then, if then, if then, we're not that.”
Grosser said no one has invented a test for product-market fit, so the question of product versus go-to-market keeps coming up. At Vercel she picked 10 large existing enterprise customers that used Vercel but not yet in front of their main site, and had the product manager and lead engineer hear all 10 in a row to build pattern recognition. Afterwards everyone agreed there were product gaps to close before the product would be much more easily sellable. She noted that some enterprises still buy before full fit, because a visionary buyer will always work through the pain.
“So we're like, great, let's go get 10 of those. Let's get the product manager, lead engineer, and let's run them through conversations with all 10 in a row, so they can get pattern recognition.”
Dan Lee said that when Nooks went all in on sales, he spent 60 to 70 percent of his week observing customers using the product, which he described as the majority of his time. The team did this by spending time on the sales floor with the sales teams using Nooks.
“Um definitely the majority like 60 70%”
Dan Lee split product sense into rigorous prioritization and taste. He described taste as empathy, putting yourself in the customer's shoes and understanding deeply what they care about. He said taste matters more in human-in-the-loop AI products because the bottleneck is quality.
“it's kind of empathy putting yourself in the customer's shoes uh understanding deeply what they care about.”
Dan Lee said it is easy to build a ton of features, so the hard part is knowing what is important. Nooks lives by a do more with less value, and he described its approach as rigorously evaluating options against each other and setting clear north-star metrics and leading indicators.
“we can do anything but not everything.”
Harry Stebbings said he had moved from ChatGPT to living in Instinct. Jack agreed the form factor draws on a fixed number of waking hours already spent online. He separated this from whether particular companies lose, since the labs may build agents themselves.
“the form factor cannibalizes. That doesn't mean that these products cannibalize the other companies, but it does mean that the form factor takes some amount of the space”
He said reference calls at $1M ARR would have shown limited use. Now customers say 'I run my whole life out of it', and the businesses are at hundreds of millions of ARR. He made the same point about coding agents and about early ChatGPT usage, and applied it to personal agents being early in their cycle.
“if you did the reference calls on their customers, when they were at a million of ARR, the And now you talk to them, and they're like, I run my whole life out of it.”
He says he runs coding agents 10 hours a day and points to lawyers using Harvey and Legora constantly as examples of all-day use. He called Instinct and Muse 'generation two' after OpenClaw, which few people could run safely. He said it is an open question, not a criticism, whether these become 'eight hours a day' apps, and that if they do, they win.
“Will we run Instinct Muse eight hours a day? If we do, I guarantee it wins, right?”
Grant Clarke says Atlas records what reps do outside the agent's handoffs and classifies those actions into buckets. Forward deployed engineers, product managers and business analysts sit 'in the boat' with renewal managers to monitor this. When a category of escalation recurs multiple times, it becomes a product requirement. The FDE team develops it in maybe two weeks, tests it for a few weeks, and puts it back into Atlas.
“The FDE team can sprint, develop that in maybe two weeks, run a few weeks of testing, and then it's right back into Atlas to handle the next customer issue that comes up.”
After the pivot, Higgsfield asked eight creative directors what was missing from AI video, and every one of them said camera control, which they called essential to storytelling. The product launched on March 31 and Mashrabov describes product-market fit as immediate. He says VFX and camera control took Higgsfield from roughly $1M to $20M ARR in about the first three months.
“We spoke to eight creative directors about their experience with AI and what's simply missing. Everyone told us that camera control does not exist in AI.”
Mashrabov says Higgsfield spent more than a year searching for a product that worked, and he takes responsibility for optimizing for hype, narrative and attention instead of building a good product. With slightly less than $5M left and feeling they had 'one attempt left', the team committed to product and PLG on the belief that the best product would win. That led to the launch that took off.
“I was so much optimizing for what's hype today, what's the right narrative, how we can hijack the attention, all these things, really. Everything instead of building a good product.”
When AJ Bruno asked whether customers would just evaluate the base models themselves, Asad Zaman said product matters. He said the intelligence may be better in one tool while the product sells better in another. He thinks product matters most in domains where institutional knowledge must be extracted and turned into workflows and integrations. He described the swing at scale: a smarter model drives usage up but pushes gross margins down, and the company then has to rebalance.
“I think there would be great application layer businesses but these are not the same sort of application layer businesses of the sort that we're used to.”
McMahon said product managers often present new demos and roadmaps without answering these questions, and reps sit there with no idea what is being said, so the whole hour is lost. He said reps care about how the product solves their customers' problems rather than what it does.
“Because they don't care what the product does. They only care about how it solves problems for their customer, who's the persona we're selling it to?”
A former solutions engineer designed the system, and CSMs use it constantly. An intake agent listens to Gong calls and monitors a Slack channel where CSMs post requests. Another agent restates the request in natural language, another checks Salesforce for whether the requester is a customer, and another logs it in a database. That gives a top-10 feature request list across customers, and requests are referenced in customers' AI transformation plans. The next step Christina Parra described as planned is an automatic alert to the customer when a request ships, to build trust.
“There's another one that checks Salesforce to see, is this a customer, is this not a customer?”
AJ said Gong's landing page now leads with 'Gong AI, ask anything' and pointed to the newly released Gong Bridge. He said this lets users find all their information across accounts quickly, rather than going to separate forecasting or call tools. He then asked Keith how Lightfield ties its own capabilities together.
“It's not about just the forecasting or the calls or whatever. It brings everything together.”
A host had observed that GTM platforms bolt features onto a left-hand menu without connected workflows. Keith agreed that everyone is racing to the roughly 100 features required to run a go-to-market team. He said making them performant and coherent together is really hard, and many players produce apps that in theory prospect, forecast and hold a CRM data model but aren't usable by a rep.
“create this Frankenstein app. You know, with a hundred features that in theory could prospect, in theory could forecast, in theory has a CRM data model, but is not a thing that a rep can use.”
The host called the distillation argument smart-sounding but divorced from how buyers value something ten times better. They said Granola could already be copied cheaply, yet it grows fast because of product team, distribution, virality or customer empathy. They added that its bot-free recording and pre-call web research come from how deeply it understands the user's workflow.
“So at the end of the day, building a company and building products is still really fucking hard.”
Asad gave this as his argument for why more functionality ends up inside each platform. He said if you charted product shipped per stage or per $1M ARR over the last ten years, it would be dramatically higher now, and he expects the trend to continue. He did not give numbers.
“Now I go into a series a product and it feels like what a series C product used to feel like”
Keith admitted Lightfield doesn't yet have the best commission planning tool. His broader point is that customers are building CPQ in Lightfield themselves, using automations, conditional fields and forms, rather than buying a separate product.
“a lot of these features are now just automations and skills that run on your system of record in the new world.”
Keith names two data-structure differences from HubSpot and Salesforce. First, the system should update itself automatically by default through APIs, which he says conventional CRMs weren't built for. Second, it should capture all interactions rather than only what fits into fields.
“I think you want to capture everything versus what's just in the fields because Like, who knows what question you're going to ask next month, you know?”
Ann says Looker's data model and semantic layer governed how data was managed and maintained, and that natural-language querying became the analytical layer on top. She cites Google's acquisition of Looker, at a price of $2.6 billion on a roughly $100 million run rate, as evidence of that value.
“it was about the data model in the semantic layer of how the data is managed and maintained”
Ann says Crunchbase launched its MCP server about a month before the recording and that the value lies in bringing data sources together through an AI interface that sits inside the seller's workflow. She says the choice of front-end AI tool, such as ChatGPT, Gemini or Grok, does not matter.
“it's about how you seamlessly bring together the data sources through the AI interface to deliver real value.”
Nerad described a CS director at another company who builds customized dashboards from product data for each customer, and said customers love it. Her concern is how that scales. She said ideas CSMs develop with customers must be fed back into the product, or the industry ends up where software was 25 years ago, with customizations it can't maintain.
“Else we're going to end up in the place where software was when I got into the industry 25 years ago with all the customizations that we aren't able to maintain too”
Matt said users of tools such as Meltwater, Cision and Muck Rack often get data that is dirty, so they export it to CSV or spreadsheets and clean thousands of rows of news articles every week, month or quarter. He said this manual work is the problem Handraise is using AI to solve.
“The data ends up being really dirty, so they have to export it to like a CSV file, Google sheets, Excel, and then they have to wrangle this data around, clean it.”
Matt described pressing a button on a narrative cluster to generate a summary in Axios-style brevity formatting that reads all the coverage and says what to know. It then looks more broadly at the news about the company and provides a 'why it matters' section. He said building something like this at TrendKite would have taken many months and still would not have hit today's quality bar.
“We're able to just press a button and it generates a summary of the narrative that gives an Axios Smart brevity style formatting on that thing”
Josh Schachter describes design-thinking user research from his time at BCG. Asking 'how often do you do X?' got thin answers like 'every other month'. Following up with 'tell me the last time that happened, give me that situation' put people in a concrete scene. He links this to London's hypothetical scenes, saying both let the person picture a situation.
“okay, great, well, tell me the last time that that happened. Like, give me that situation.”
Christopher says there is no need for an analyst to quantify the problem or a designer to mock up options. Someone on the phones and someone in product sit together over coffee, find something, verify its impact and ship it. In his office, four or five people may talk for three hours around couches over data and customer feedback, and by the end the software is written.
“You need somebody on the phone and somebody in product to sit together over a coffee and get a lead on something you could do, verify that it's going to have this impact and ship it 45 minutes later.”