This is a timing post, not a thesis post. The thesis lives in describe outcome, not workflow,workflow editors are the wrong abstraction once intelligence has moved into the system. This post is the narrower question: why now? What changed in 2026 that makes the bet against workflow platforms reasonable today and would have been reckless a year earlier?

Three signals lined up. Each one is suggestive on its own; together they mark a shift in market share that won't reverse without something equally large pointing the other way. Workflow platforms are not going away,they are losing the general-operator segment, which was their growth engine, to outcome-based agents.

The bet, stated plainly

Workflow platforms,Zapier, Make, n8n, Pipedream, Activepieces, Workato, Tray, Relay, IFTTT,are losing share in 2026 to outcome-based AI agents in the general-operator segment. They will retain share in regulated, irreversible-action, and compliance-heavy categories. The growth segment shifts; the surviving segment is smaller and slower-growing than the segment they're losing.

This is not a "Zapier is dying" claim. Zapier is profitable, established, and well-positioned in its surviving segment. The claim is narrower: the share of operator automation that flows to workflow tools versus to outcome-based agents tilts toward agents through 2026 and into 2027. The compounding goes the other way from where it went 2014-2024.

The three signals that lined up in 2026

Signal 1,model reliability reached the bar. Through 2024 and 2025, foundation models gained the multi-step tool-use reliability needed to run autonomous tasks for someone else. The line that mattered was not "smartest model" but "reliable enough that humans don't have to babysit". 2024 was below that line for most categories; 2025 and early 2026 crossed it for inbox triage, lead follow-up, KPI reports, watch lists, and competitor tracking. The reliability bar matters more than the capability bar,a flashier model that doesn't reliably finish doesn't replace a workflow tool.

Signal 2,inference cost dropped to commercial-viable. Per-token prices on flagship models dropped sharply over 2023-2025 across leading vendors,visible by comparing snapshots of OpenAI and Anthropic API pricing pages over time, plus the new tiers introduced for cheaper, faster models. That's the difference between an outcome-based agent that runs at a $50 monthly subscription and one that runs at $500. The price the agent can charge has to clear the cost-of-inference, and the cost-of-inference cleared the relevant price band in 2025. The unit-economics math walks through the implications.

Signal 3,ChatGPT Workspace Agents launched April 2026. When OpenAI ships an outcome-based-agent surface (OpenAI, ChatGPT agents documentation, accessed 2026-05-05), they validate the category at a scale no startup can. The launch did not invent the category,agent-based automation was already shipping at multiple startups,but it confirmed that the largest distribution surface in AI is committed to outcome-based, not workflow-based, automation. Validation from the largest player is a market-share signal, not a product signal.

Three lines crossed the threshold in 2024-2026 2023 2024 2025 2026 Commercial-viable threshold Model reliability Inference cost (inverted) Category validation Stylised. Captures the qualitative trend across reliability benchmarks, model API pricing, and category-defining product launches.
Three lines crossed the threshold in 2024-2026. None individually is sufficient; collectively they mark the shift.

Where the share goes

The agentic AI market is projected to grow from approximately $5.4 billion in 2024 to $93.3 billion by 2034,a compound annual growth rate above 32% (Precedence Research, agentic AI market, retrieved 2026-05-05). The bulk of that growth is general-operator automation moving from workflow tools to outcome-based agents.

The split inside the agentic-AI growth bucket is harder to forecast precisely, but the directional logic is straightforward: every recurring task currently automated through a workflow tool is a candidate for an outcome-based agent if the agent can do it reliably and cheaply enough. The reliability and cost gates closed in 2025-2026; the migration starts now.

For Gravity, this is the structural argument for being a deployer of outcome-based agents now rather than five years ago or two years from now. The market is large, the timing is correct, and the competitive landscape,covered in detail in upcoming comparison posts,has not yet consolidated around a winner.

What survives in the workflow-tool category

The categories that survive: regulated workflows, irreversible-action workflows, audit-heavy compliance workflows. Financial reconciliation that touches general ledger postings. Healthcare workflows with compliance audit trails. Anything that has to be auditable step-by-step in a way an agent's reasoning trace cannot easily satisfy.

Workflow tools that focus on these segments will continue to grow. They will grow more slowly than the broader agentic market, but they will not disappear. The bet against workflow platforms is a bet against the broader-segment growth, not against the category's existence.

This matters for incumbents. Zapier-the-business is well-positioned for the surviving segment because it has the integration breadth and reliability story regulated buyers want. n8n is well-positioned for self-hosted, on-prem, regulated deployments. Make and the rest will fight harder for the same shrinking pie. The strategic question for incumbents is whether to defend or migrate; both are credible plays.

The "you're early" risk

The biggest risk on this bet is being early. Early can mean "before the wave" or "before the wave decides not to come". Yara AI was early on AI mental health and shut down (Fortune, November 2025). Vibe AI was early on AI companions and shut down. The market does not reward early; it rewards right-sized for the wave that actually arrives.

The signals above are the reason this bet does not feel like Yara AI's bet. Reliability has crossed the bar; cost has crossed the band; the largest distribution surface has validated the category. None of those were true for AI mental health in 2025. They are true for outcome-based agents in 2026. The framework that made bet four feel different is in the three checks I missed; the timing is the part the framework cannot guarantee.

Honest disclosure

I am building Gravity, an outcome-based agent platform. The bet against workflow platforms benefits Gravity; that's a conflict of interest, not a hidden one. The argument should be evaluated on its own merits,the three signals are real or they aren't, and the timing is right or it isn't, regardless of who profits from being right.

I'm wrong on a third bet. The framework that emerged after that third bet is what makes this fourth one feel different. If you have a strong prior that workflow platforms will hold the general-operator segment longer than I think, I want to hear the case,my email is at the top of /contact.

Frequently asked questions

Are workflow platforms going away?

Workflow platforms are not going away,they are losing share in the everyday-operator segment to outcome-based AI agents. They are likely to survive in regulated, irreversible, and audit-heavy categories where step-level control is the right abstraction. The bet is on a structural shift in market share, not extinction.

Why is 2026 the inflection point?

Three signals lined up in early 2026. Foundation models reached the reliability bar required to run multi-step tasks autonomously. Compute prices dropped enough to make outcome-based agents commercially viable. ChatGPT Workspace Agents launched in April 2026, validating outcome-based automation as a category. Each signal alone is suggestive; together they mark the inflection.

Which workflow platforms survive the shift?

Workflow platforms that survive the shift will be the ones serving high-stakes, regulated, or irreversible-action workflows where step-level control is genuinely required,financial reconciliation, healthcare workflows with compliance audit trails, regulated industry processes. Workflow tools targeting general-operator automation are the ones losing share.

How big is the AI agent market in 2026?

The agentic AI market is projected to grow from approximately $5.4 billion in 2024 to $93.3 billion by 2034, a compound annual growth rate above 32% according to Precedence Research. The growth is driven by general-operator automation moving from workflow tools to outcome-based agents,the same shift this post argues.

What is the risk of being wrong about this timing?

The risk is being early. Yara AI was early on AI mental health and shut down. Vibe AI was early on AI companions and shut down. Early can mean "before the wave" or "before the wave decides not to come". The bet against workflow platforms is the same kind of timing call, and the framework that made bet four feel different is in the three-checks post.

Three takeaways before you close this tab

Sources