AI-First Isn't a Feature.It's a New Operating Model
Most SaaS homepages now say AI-First. Most of them mean an AI feature. AI-First is how a company approaches work, and for a growth team it shows up in the distance between what a user wants and the outcome they get.
Panel recap. November 18, 2026, hosted by Foldspace. With Sunil Subhedar (Former Growth Marketing Lead, Anthropic · AI GTM Advisor), Jakub Tutaj (Staff Product Manager, Monetization, Mixmax) and Allan Mørch (CEO & Founder, AskCody). Moderated by Mickey Alon (CEO & Co-Founder, Foldspace).
Building a feature used to take an engineering team months. Now it can take hours. When building gets that cheap, the hard part doesn't go away. It moves from whether you can build something to what you should build and how you should operate.
Start with how the work gets done
Buy AI licenses, bolt them onto the playbooks you already run, and everything gets faster, including the parts that never worked. The alternative is to design the work as if you were starting today. At AskCody, that meant organizing around shared intelligence instead of an org chart: every playbook, procedure and piece of company context became a shared skill that everyone's agents run on. What one person teaches the system, the next person gets.
Sunil puts it in three parts: the people, the process, and the harness and context you build around the model. In his experience, most of what makes it work happens around the model, not in it. For a product team, it starts with the core job customers come for, not with every feature. Count the clicks that job takes today, the click tax, and ask how many of them AI can take off.
Sign-up: buyers ask an AI first
More buyers ask an AI assistant to do their research before they reach your website. Sunil's advice is to find out what those assistants say about you, because it can differ from what your site says.
Standing out is harder when everyone claims to solve the same problem. He sees three things that still set a product apart: proprietary data built on real domain expertise, a complex process where errors aren't an option and you've built the harness to run it, or a last mile you deliver far faster than a general model could.
The pitch changes too. Sunil has seen almost an aversion to the word AI in a pitch. Lead with it and you get compared to a frontier lab, so lead with the outcome and the KPIs. He calls it demos, not memos: if there's no aha moment in the first five minutes, change the pitch.
Testing got cheap. A homepage A/B test used to mean duplicating the page, splitting traffic and a round of debate. At Mixmax, an idea can now be live the same day. When a test costs that little, deciding what to test is the real work.
Activation: let users say what they came to do
Onboarding has long been built on a guess pieced together from support tickets and sales calls. Mixmax stopped guessing. In lightweight pilots, new users could simply ask for what they wanted to accomplish. Those users engaged much more and were much more likely to convert. Before trusting it, the team checked whether these were just the high-intent users, and found no sign of it. Mixmax's published result: users who engaged the Product Agent showed a 4.4x conversion lift, confirmed with three anti-bias tests against matched historical cohorts.
Each request also showed the team whether users got what they came for. When users were blocked on something the product already does, the help articles got fixed. Mixmax is now on the next loop: the product does it for them, so users no longer have to learn the UI to get the result.
Jakub's metrics mostly haven't changed: whether users engage, activate and stay. If he had to add one AI metric, it's how often users have AI-assisted touches in the product, and whether that number grows.
Revenue: customers ask for outcomes
Customers ask for what they can imagine: a blue button, a better report. When you can build almost anything, the job is to hear the outcome underneath the request. AskCody launched an in-product assistant, built on Foldspace, for onboarding and guidance. It first showed them the gaps in their help center. Then users started asking for outcomes without realizing it, and Allan called it the best feedback loop they've ever had built into the product.
As Allan put it, being able to build everything doesn't mean you can distribute it. The commercial questions haven't changed: will customers use it, and does it create value you can capture? The same holds for AI costs. The first question isn't how much to spend but whether you have product-market fit. Without it, spend what it takes to get back to it. With it, ask customers what the next thing is worth to them.
Production isn't the bottleneck anymore. Decisions are.
Agents run around the clock, and a team can now produce far more than its value chain was built for. As Allan sees it, nobody can decide at that pace, and the decision model is what slows a company down. Jakub's rule helps: if a decision is easy to reverse, test and learn fast; if it isn't, align the stakeholders first. Even then it moves faster than it used to. His first packaging update at Mixmax took almost a year from idea to launch. The second took about three months.
What to take back to your funnel
Close the distance between what users want and what they get at every stage, and keep your decisions moving as fast as your team can now build.
If you want to see what that looks like inside your own product, start a free trial.
Mickey Alon 

