Arseny Shatokhin

@tim_yakubson stack

I cancelled my Clay subscription, and I run a Clay agency: My 2025 outbound stack cost $3,300 a month. My 2026 stack costs $700, and it still does the same sourcing, enrichment and filtering without touching a paid Clay credit anymore. A year ago that wouldn't have made sense. Now Claude Code quietly handles the day-to-day workflows I used to open Clay for, which raises an uncomfortable question: how many people are paying full Clay pricing out of pure habit, not because the work needs it? To be clear, this is NOT me saying Clay is dead. It still does jobs nothing else on the market does. The role it plays in your stack is just shrinking. I recorded a full breakdown of the 2026 stack, and I'm giving it away free, as one of the first UK-certified Clay experts telling you to de-risk Clay. Comment "STACK" and here's everything you get, free: • The $3,300 to $700 swap, line by line, so you see exactly what moved off Clay • The 4 workflows Claude Code now runs (sourcing, enrichment, filtering, the daily grind) • The jobs Clay still does better than anything, where you keep paying • The orchestration setup that lets you offload more than you'd expect • The Free Plan config I actually run my agency on PS I profit from people staying on Clay. I'm still telling you to shrink the bill.

Most teams try to automate processes they can't even explain. Don't automate chaos. Document, then automate.

These four things map directly to the three components of agents: → workflows become instructions → documentation becomes knowledge → communication channels become integration points

If the process isn't documented, don't build your agent. Document first. Automate second. Here's what we need before building an agent: (a thread) 🧵

We've never built a perfect agent on the first try. We've built 100s of agents that got better every week. Iteration isn't optional. It's the only way agents survive in production.

This works because you can't predict edge cases in a test environment. Client needs change as they use the agent. And real data beats your assumptions every time.

Agent development isn't a waterfall project. It's an endless loop. Here's the process that actually works after building agents for 50+ companies: (a thread) 🧵

Most people think building the agent is the hard part. It's not. In our agency, it takes 2-3 days to build an agent. Then it takes another 3 days to deploy it. Deployment is the part nobody talks about. You have to: – connect APIs – set up webhooks – integrate with tools – configure auth flows – write the documentation And if it's not convenient to use, it doesn't matter how good the agent is. Your agent needs to work where users already are - not in some separate app they'll never open. Most teams spend all their time perfecting the agent and wonder why nobody uses it. But the teams that win treat deployment as seriously as development.