🤖 No-code is becoming code under the hood of AI agents: Zapier's playbook

👤 Wade Foster - co-founder and CEO of Zapier (web app integration and automation platform)
🎤 Nathan Labenz - host of The Cognitive Revolution (podcast exploring the impact of AI)

📺 Cognitive Revolution – 👥 17.7K subscribers
⏱ 1h 7m
🗓 18.09.26 (2d ago)
👁 6.6K

📝 Wade Foster explains why 80% of agent tasks should rely on deterministic code, why per-seat pricing is dying, and how Zapier is turning its tools into headless backends for third-party AI environments.

━━━━━━━━━━━━━━━

💡 KEY TAKEAWAYS

1️⃣ 80% of agent tasks require deterministic code, not reasoning
Zapier data reveals that AI reasoning is overkill and unreliable for 80% of workflow steps. The most robust architecture uses agents solely for non-deterministic branching while delegating routine execution to rigid, deterministic code.

2️⃣ Top model scores only 40% on the AutomationBench benchmark
Zapier's benchmark of 600 real-world business tasks exposes the limits of current LLMs, with the top-performing model achieving just a 40% success rate. Fully outsourcing complex workflows to AI without guardrails remains impractical.

3️⃣ Users are anchoring in primary AI environments
Knowledge workers are settling into single daily drivers like Cursor, Claude Code, or ChatGPT. Platforms like Zapier must adapt via MCP (Model Context Protocol) to serve as headless backends inside these chosen interfaces.

4️⃣ Traditional no-code is dying as agents write code directly
Manually stringing together visual blocks is becoming obsolete. Users describe their intent in natural language, the agent writes the underlying code, and the human simply reviews the resulting visual diagram.

5️⃣ Self-healing workflows powered by multi-agent consensus
For internal support, Zapier spins up five parallel agents to diagnose errors. If at least four reach consensus on the root cause, the system automatically fixes the pipeline without human engineer intervention.

6️⃣ Per-seat SaaS pricing gives way to usage-based models
As AI agents generate the bulk of business value rather than human seat holders, SaaS pricing models are rapidly shifting toward token consumption and outcome-based billing.

7️⃣ Radical communication transparency supercharges internal AI agents
Zapier shifted most team communication into public Slack channels. This provides internal AI agents with full organizational context, drastically reducing hallucinations and outdated outputs.

💬 «The new no-code is code. You really only want the AI to reason over the things that you need it to reason for.»

🎯 WHAT TO DO WITH THIS
Decouple your workflows into deterministic code and LLM reasoning: never deploy AI where a simple script gets the job done.

💵 0.11$

@icereadspodcasts