No Code Is Code: Zapier CEO Wade Foster on Headless Tools, Zapier MCP & Automation Bench

Episode Summary

Executive Summary: Wade Foster argues Zapier is evolving from a no-code automation tool into a headless orchestration layer that plugs into users’ preferred AI daily drivers. He says models still struggle on business workflows, so deterministic code plus AI is the best near-term pattern, while product value will increasingly come from context, retrieval, troubleshooting loops, security, and making AI visibly useful inside real work.

Main Topics: AI daily drivers and the new application layer (Priority: 5/5): Foster says most users now settle on one primary AI environment (e.g., Cursor, Claude, ChatGPT) and that tools like Zapier must work inside those environments rather than compete to be the main interface. Model limits, benchmarks, and deterministic workflows (Priority: 5/5): He describes Zapier’s automation benchmark, says top models still only solve about 40% of tasks accurately, and argues many agentic tasks should actually be done by code with AI reserved for reasoning and exception handling. AI-assisted workflow discovery and productization (Priority: 4/5): Zapier is building systems that watch a user’s work across Gmail, Slack, browser, and chat to suggest automations and even productize these recommendations for customers. Self-improving operational loops and multi-agent evaluation (Priority: 4/5): Foster explains how Zapier uses multiple models/agents, human thumbs-up/down feedback, and repeated evaluation to improve workflow reliability and troubleshooting. Org design, public-by-default communication, and AI fluency (Priority: 4/5): He discusses how AI adoption changed Zapier’s org structure, why the chief people officer led AI transformation at that stage, and why more public communication improves both human coordination and AI usefulness. Security, credentials, and early access to frontier models (Priority: 4/5): Given Zapier’s role handling many credentials, Foster emphasizes strong stewardship, using frontier models for both offense and defense, and wanting early access to cutting-edge capabilities. Pricing, competition, and co-authorship norms (Priority: 3/5): He argues seat-based pricing is fading, usage/outcome pricing will dominate, competition is more about educating the mass market than winning head-to-head with model labs, and AI should assist writing without removing human accountability.

Key Arguments: Most people are converging on a single AI 'daily driver,' so software vendors need to integrate into that environment instead of forcing users onto a new platform. Frontier models are improving, but for many business workflows they still fall short; Zapier’s benchmark shows ample room for improvement and confirms the value of orchestration. A large share of agentic work should be deterministic code, not AI; AI should handle reasoning, planning, troubleshooting, and edge cases. AI can now do a lot of the build work for automation, making classic no-code interfaces less central; visualization and human verification remain important. Product differentiation will come from proprietary workflow data, feedback loops, and context from customer systems, not just from model access alone. Public-by-default communication helps both people and AI agents understand company context, but sensitive areas like HR incidents and active security issues still need private channels. Security is becoming more important because modern models can find vulnerabilities faster, but they can also help fix them faster. Pricing models are shifting from seats toward usage or outcomes because intelligence is becoming embedded into the core product experience. AI can improve writing and communication, but employees must still understand, verify, and stand behind what they send. The biggest market opportunity remains helping the broader population understand what AI can do for them; most people still use only a tiny fraction of available capability.

Data Points: Customers using Zapier to delegate tasks to AI: 400,000 - Number cited from the prior interview timeframe; customers had delegated work to AI via Zapier. Tasks delegated to AI via Zapier: 100 million+ - Tasks already delegated by customers when Wade was last on the show. Automation Bench size: ~600 tasks - Zapier benchmark covering knowledge-work workflows across marketing, sales, HR, and operations. Top benchmark performance mentioned: ~40% accuracy - Wade said Astra/GPT-6 was the new state of the art and achieved about 40% task accuracy on Zapier’s Automation Bench. Relative cost comparison: Gemini 3.7 at a fraction of the cost - Used as an example of a cheaper model that performs well though not necessarily at the same level as the top model. Human usage of AI internally: ~100% of employees within about a year of ChatGPT launch - Zapier got nearly all employees using AI day to day relatively quickly. Agent troubleshooting setup: 5 agents - Zapier’s auto-email program reportedly spins up five independent agents to evaluate issues. Agreement threshold for troubleshooting: 4 of 5 agents agree - If four of five agents agree, Zapier treats it as a strong signal of the true issue. Individual token spending example: $30,000/month - Wade referenced some engineers spending this much monthly on tokens at Zapier. Average Athena client time saved: 15+ hours/week - Sponsor mention in transcript about Athena clients. Mercury customer scale: 300,000+ - Sponsor mention in transcript about Mercury. Company size: ~800 employees - Wade referenced Zapier as having an almost 800-person organization.

Pivotal Quotes: "Most folks seem to have adopted their own daily AI driver tool." — Wade Foster: Explaining the emerging AI product pattern and why Zapier must integrate into users’ preferred harnesses rather than replace them. "The idea of building a no-code, it feels antiquated to me. It's like the new no-code is code." — Wade Foster: Describing how AI now builds many workflows faster than humans, while deterministic code remains the execution layer. "I have no problems with people using AI for communication at Zapier. What I really have a problem with is low-quality communications." — Wade Foster: Clarifying Zapier’s stance on AI-assisted writing, emphasizing quality and accountability over tool purity.

Implications: AI adoption is moving from novelty to embedded workflow design. Winners will combine model access, proprietary context, automation infrastructure, and strong governance—while helping ordinary users discover practical uses beyond chat.