IA40 Summit 2026 recap: AI agents and hardware NPD
News
Our CEO, Ford Davidson, spent September 30th at Madrona's sixth annual IA40 Summit in Seattle.
Our CEO, Ford, spent September 30th at Madrona's sixth annual IA40 Summit in Seattle, alongside leaders from Microsoft, Anthropic, AWS, Goldman Sachs, Stripe, Figma, Atlassian, and McKinsey. The topic running through nearly every session was the same: what actually changes, in a business and not just in a demo, once AI agents start doing real work instead of just assisting with it.
A few of those conversations are worth unpacking in detail, because they point to decisions every product organization, including hardware teams, will be making over the next year.
The gap between adoption and impact
McKinsey opened its briefing with new data from its report, "The State of AI in 2026: On the Road to ROI" (based on a survey of 1,719 organizations). Adoption is no longer the story: 89% of organizations now use AI in at least one function, up from just 20% in 2017. The story is that payoff hasn't followed. Only 37% of organizations report any positive impact on EBIT from AI, a number that hasn't moved since last year.
That gap matters more than either number alone. It means most organizations have absorbed the cost and complexity of AI adoption without yet seeing it show up on the income statement, which is exactly the position a lot of product and operations leaders find themselves defending internally right now.
What the 6% who do see results have in common
McKinsey's data also isolated the 6% of organizations that attribute 5% or more of EBIT to AI. The defining trait of that group isn't more AI spend or a flashier tool. It's that 75% of them had fundamentally redesigned a workflow because of AI, compared to only 25% a year earlier. They rebuilt the process around what agents can do, rather than inserting an AI step into a process built for people.
That distinction came up again on the main stage, in a line from the "Software's Agentic Rewrite" panel that's worth quoting directly: "Why automate a workflow that is going away? Reimagine it." The high performers aren't automating their old workflow faster. They're asking whether that workflow should still exist in its current form at all.
Agents are becoming users of software, not just a way to build it
A theme that surfaced across several sessions, from Microsoft's Charles Lamanna to Anthropic CTO Rahul Patil, is that agents are starting to use software directly, not just help people use it faster. Figma reported that two-thirds of its monthly active users are no longer designers, a shift driven by Figma Make's prompt-to-prototype agentic flow. Vercel reported that 56% of its deployments are now carried out by agents rather than engineers. Jean-Denis Greze, CEO of Town, put it bluntly: as agent capability keeps climbing, "everything is going to become a thin app" sitting on top of an agent that does the actual work.
The practical implication is that "who is the user" is no longer a settled question for any product team. An increasing share of the people, or systems, interacting with a piece of software will be agents acting on a human's behalf, often asynchronously and outside a traditional UI session.
The question nobody on stage could answer
The most interesting moment of the day, by GeekWire's own account, wasn't a confident prediction. It was Rahul Patil declining to answer a direct question about who owns the data an AI agent generates while working on a company's behalf. His answer was that companies "will use every data that's available" to improve their systems, which is a true statement and also not an answer.
That ambiguity isn't academic. Stripe's Maia Josebachvili described agent commerce on its platform sitting flat for eight to nine months and then surging sharply in six weeks, a shift that's already creating friction: merchants are discovering that agent-mediated purchases can mean losing the upsell and advertising relationship they used to have directly with a customer. Amazon blocking Meta's Muse assistant from shopping on its platform last month is the same tension playing out in public. Whoever sits between a company and its customer, model provider or platform, is now fighting over a relationship that used to be uncontested.
What this means for hardware product teams
It's worth being direct about one thing: IA40's speaker lineup was overwhelmingly software and fintech (Stripe, Vercel, Figma, and consumer AI products), not hardware-native. Nothing above was said from that stage about hardware new product development specifically. What follows is our own read, applying those patterns to a world we know well, not a claim anyone at IA40 made.
Applied to hardware NPD, the pattern holds up. Stage-gate and NPI processes are exactly the kind of workflow that gets automated one checklist step at a time rather than reimagined, and they're exactly the kind of process where "who owns the record of what happened" determines who agents end up trusting. Our own takeaway, and the thing we're building toward, is to make sure Enzzo is that trusted system of record: a place agents can read from and write to through a clean, governed interface like MCP, not a UI that gets quietly routed around as agents take on more of the work.
That also means designing for a growing share of usage that isn't a human clicking through a screen, and building the kind of feedback loop that lets a customer see their own process actually improving over time, the same habit McKinsey's high performers already practice.
The open question for your team
McKinsey's math is sobering on its own: by their estimate, the roughly $7-8 trillion of AI infrastructure spending underway needs to generate $3.5-5 trillion of net new value every year to justify itself at a reasonable return. That's not a small bet, and it's not one any single company settles by attending a summit.
The more useful question is a narrower one, and it's the one we keep coming back to: which part of your product development process would you trust an agent to touch today, and how would you know if it started quietly getting it wrong? Most teams haven't had to answer that yet. Ours is one of the ones trying to.
Get Started with Enzzo
Contact Sales