Web Apps Are Toast.

Bret Taylor is in a position almost no one else occupies right now: building one of the fastest-growing enterprise AI companies (Sierra), while serving as board chair of the most consequential AI lab in the world (OpenAI). That dual vantage point came through in every answer at HumanX 2026 San Francisco. This wasn't a founder selling a vision. It was someone describing what he's already watching happen.

Here's what stood out.

The Web App Is, In His Words, "Broadly Toast"

Taylor didn't bury the lede. When asked whether web apps are finished, he said yes — with caveats, but not many.

His argument follows the arc of computing history. In the 80s and 90s, the big unlock was storing company data in a database. Then the internet arrived and we built web apps: forms, fields, and buttons that let humans edit rows in that database. SaaS was essentially a 30-year exercise in making that interaction slightly less painful.

Now comes the agent layer. Instead of logging into a system you use twice a year, learning its navigation, clicking through dropdowns — you just ask. You tell an agent to carve sales territories, onboard a vendor, or close the books. It does the work. You review the trade-offs.

"Will you be in a web app, clicking buttons, looking at visualizations," Taylor asked the HumanX audience, "or will you ask and say, 'why don't you carve the territories for my salespeople and show me the trade-offs?'"

The answer, he argued, is obvious And it changes the architecture of software in ways the incumbent SaaS industry hasn't fully reckoned with. Every system of record effectively becomes "headless." The interface disappears. What's left is the logic, the workflows, the guardrails — and an agent to navigate them.

Applied AI Is the Real Prize — and Nobody's Building It Yet

Taylor is vocal about something he thinks gets overlooked in the AI discourse: the applied AI market.

The current conversation is dominated by model benchmarks, infrastructure spend, and which labs are winning the capabilities race. Taylor's thesis is that most of the eventual value will be unlocked somewhere else entirely — in companies that use AI to solve specific, boring, expensive business problems. Not "AI" as a product. AI as a solution to a problem a CFO or a COO actually has.

He pointed to financial auditing as an example. Close a quarter and you enter a multi-week period of humans reviewing numbers, reconciling revenue recognition rules, and coordinating with outside auditors. "Not only could an AI agent do that for much less money and maybe shorten the time between quarter close — you're going to find inconsistencies," he said. "What a great use of technology."

The problem? Almost nobody is building it. The Venn diagram of people who can build agents and understand accounting rules is, in his words, "the empty set."

That gap is where he thinks the next wave of enterprise AI value lives. Not in the model. Not in the infrastructure. In the person who actually knows the domain and is willing to build for it.

Sierra's Bet: Complexity Is the Moat

With Ghostwriter — Sierra's recently launched agent that builds other agents — Taylor is making a specific product bet. Not just that enterprises want agents, but that deploying them successfully at scale requires more than good software. It requires a trusted advisor who understands compliance, change management, and what it means to go live at a Fortune 20 healthcare payer.

Sierra went live with Cigna in two months. With Nordstrom in four weeks. Taylor spent time at HumanX unpacking why those timelines were possible — and it wasn't just engineering. It was the organizational navigation that came with it: EU regulations, local AI compliance requirements, 40 call centers being consolidated into one agent.

"I would like to be the company that the largest companies in the world come to when they want to be successful with AI," Taylor said. And his bet is that success, at that scale, isn't just a model problem. It's an implementation problem. A change management problem. A problem that requires people who have done it before.

On OpenAI and the Layers That Matter

Asked how he manages the obvious tension between running Sierra and chairing OpenAI — two entities with some competitive surface area — Taylor was direct about how he thinks about the market structure.

"There are very clear layers of the value stack," he said. The models themselves. The agent harnesses built around them (Codex, Claude). And then the applied AI market — companies like Sierra in customer experience, Harvey in legal — that aren't selling AI at all. They're selling solutions to business problems.

His view: the applied AI market is durable precisely because most businesses don't want to build software. They want their problems solved. And as the marginal cost of building software continues to fall, that dynamic doesn't change — it intensifies.

What This Means If You're Running a Business

Taylor's framing at HumanX wasn't abstract. It was a roadmap for how enterprise leaders should be thinking right now:

The technology isn't the hard part anymore. Deploying it successfully — across teams, workflows, compliance requirements, change-resistant organizations — is. The executives who will win this decade are the ones who stop looking for a software product that solves their AI problem, and start finding partners who can help them actually go live.

The web app had a good run. The agent era has started. The only real question is whether your company is building toward it or waiting to be asked.

Watch the full session on-demand. 

Register for HumanX 2026 Amsterdam (September 22–24, 2026) or HumanX 2027 Las Vegas (March 7–10, 2027).