Wade Foster on how Zapier scaled profitably without raising another round — and why agentic AI is still too unreliable for complex workflows

Summary

Zapier reached profitability in 2014 — two years after its YC batch — and has not raised outside capital since its seed round. Wade Foster says that was never a rigid ideological stance. The goal was to find a business model sound enough that raising more money never became the urgent answer. Every time Zapier ran the exercise of identifying its growth bottleneck, the answer came back as management bandwidth or a missing product feature, not capital.

Foster pushes back gently on the seedstrapping framing that has become fashionable. Founders who frame it as 'we don't want to raise more money' are often really saying they don't want more dilution, which is the wrong lens. The right question is whether money is actually the constraint — and for most companies at most moments, it isn't.

The business held its discipline partly because of where it was built. All three co-founders came out of Veterans United, a Missouri mortgage company owned by two brothers who never raised outside capital and scaled from 500 to 1,000 employees during the financial crisis. Mailchimp, one of Zapier's earliest and most successful integration partners, was entirely bootstrapped. Those were the reference points — not the Reid Hoffman blitzscaling playbook that dominated Silicon Valley at the time. When a VC told Foster 'no important company has ever been built this way,' the counterfactuals were already in front of him.

AI in the product

Zapier has run two parallel AI bets: a standalone agentic product called Zapier Agents, and AI steps injected into its classic deterministic workflow builder. The workflow-plus-AI-steps approach has been far more successful. Foster's explanation is direct — at a 10% error rate per step, an agent running 10 or 20 tasks in sequence compounds its mistakes badly enough to produce useless output. Wrapping AI inside a deterministic harness, and deploying it only at the steps where it genuinely adds value — summarisation, sentiment analysis, email generation — delivers reliability that pure agents currently can't match.

That model has generated more than 250 million tasks run through Zapier with AI steps embedded. Foster still believes fully agentic workflows are the eventual destination; he just doesn't think reliability will get there as fast as the market expects.

Zapier began experimenting with generative AI a few months before ChatGPT launched, and when GPT-4 arrived, the team briefly wondered whether the classic workflow product was durable at all. What they watched instead was AI usage on the traditional workflow builder grow consistently, which reoriented the product roadmap toward helping customers move gradually from deterministic workflows toward agentic ones, mixing the two as reliability improves.

Pricing

Zapier has charged per task since the beginning — what Foster describes as pricing based on work done. That structure predates the current 'outcome-based pricing' conversation by more than a decade. The tension it creates for a horizontal product is that high task volume doesn't reliably mean high customer value, and low volume doesn't mean low value. A vertical agent selling into customer support can do clean math: this many tickets closed, at this price per ticket, against a human rep's fully loaded annual cost. A horizontal platform serving GTM, legal, finance, and marketing teams simultaneously can't price that cleanly. Foster is trying to abstract the complexity away entirely — one task price regardless of whether AI is involved — and absorb the model cost variance internally.

Where new demand is coming from

Two customer segments have moved faster than Foster expected. Industries heavy in PDFs and paper documentation — where the SaaS stack was always clunky because it required structured data — are adopting quickly because LLMs are unusually good at turning unstructured documents into structured outputs. Media companies have also moved faster than many tech companies, which Foster attributes to having been on the wrong side of internet disruption the first time around.

The automation engineer

A recognisable persona has existed inside Zapier's customer base for 15 years — the person who kept the integration plumbing running, often underappreciated. That role is now getting formal recognition inside organisations as the 'AI automation engineer.' Foster says Zapier is seeing this cottage industry grow around implementation, and the most effective version combines systems-thinking capability with domain expertise: an HR or legal professional who also knows how to build workflows can move faster and create more value than a generalist.

MCP

Foster describes MCP (Model Context Protocol) as a protocol for agent-to-agent communication, analogous to HTTP or REST. The most concrete use case he sees today is internal tooling built inside Claude projects — a sales team, for example, connecting HubSpot, ZoomInfo, and a research tool to generate pre-meeting briefs on demand. He views it as early innings and is honest that the broader use-case landscape hasn't materialised yet. For high-volume, repeated tasks, a direct API call and deterministic code wins on speed, cost, and reliability. MCP earns its place when the task requires navigating multiple data sources fluidly rather than hammering a single endpoint.

On AI and jobs

Foster's view is that capitalism tends to absorb productivity gains through competition rather than permanently concentrating them as margin. If software companies see costs fall and margins rise, competitors invest more in customer acquisition — which itself employs people. The shift he does acknowledge is compositional: the skills required to be employable are changing, and the new-grad employment data beginning to surface suggests that adjustment is already underway. What he sees in Zapier's own customer base is small businesses using automation to do work they could never have staffed for, generating more revenue, and then reinvesting a portion of that in people.