Ryan Carson: "I Spent $20,000 in One Month on Devin — Here's What I Learned"

Published this week on Lenny's Newsletter, the experience shared by Ryan Carson — a five-time founder (Treehouse, Carsonified…) — is making waves in the global tech sphere. The title alone explains why: "I Spent $20,000 in One Month on Devin. Here's What I Learned."

Devin is the autonomous coding AI agent from Cognition Labs, capable of taking a ticket, coding the feature, testing, committing, and opening a PR — all without human intervention. Billed per use, costs can quickly add up. Carson is running 15 in parallel to build his new startup solo, without hiring a single engineer.

How He Uses His 15 Devins

The principle: each Devin instance specializes in a product area (frontend, backend, database, email integration, tests, docs…). Carson acts as project manager: he writes tickets, allocates Devins, reviews PRs, and resolves conflicts. In a typical week:

• He spends ~40 hours writing clear and detailed tickets
• He reviews ~120 pull requests opened by his agents
• He rejects about 40% of the proposals as insufficient and sends them back with comments
• He merges the others after testing

Result: the product velocity of a team of 4-5 seniors, with only one person in the decision-making loop.

The 3-Step Workflow

Carson also shared — via the podcast How I AI — his methodology based on his experience:

1. Cursor for spec-work. He uses Cursor (the AI-native editor) to write ultra-detailed, nearly executable specifications. "Vibe coding," but disciplined.
2. Devin for execution. The specifications are then given to Devin as the ticket's entry point. Devin self-organizes.
3. Codex + OpenClaw for review and orchestration. A third agent reviews Devin's PRs before Carson touches them.

It's a pipeline of three different AI brains, each in its role. The chain is expensive, but it delivers.

At $20,000/month for agents, Carson estimates he's replacing about $200,000-$300,000/month in senior engineer salaries. The ROI is stark.

What Works

• Velocity. On well-scoped features, Devin delivers in hours what a human would deliver in days.
• Availability. No agent asks for leave, gets demoralized, or leaves for a competitor.
• Debriefing. Each Devin explains in detail what it did, which choices it discarded, and why. Better documentation than a standard human team.

What Doesn't Work — or Stalls

Carson is candid about the limitations:

• Ambiguous tasks. On a poorly framed ticket or a vague business problem, Devin goes in strange directions. The quality of the spec determines everything.
• Architectural consistency. Without a human maintaining an overall vision, the 15 Devins can introduce divergences (inconsistent frameworks, duplications). Carson spends a significant amount of time re-architecting.
• Regressions. Devin can break working code with confidence. Massive automated testing is crucial.
• Cost. $20,000/month is the price of a senior US engineer. For a pre-revenue startup, the cash burn is real.

What It Reveals About the Future of Dev Work

Three evolutions:

1. The solo tech founder becomes viable. What required 3-5 engineers now needs 1 person + 15 agents. Micro-startups with a single human will proliferate (see also the doubling of Stripe creations attributed to AI).
2. The real skill becomes specification clarity. Writing an ultra-precise ticket is more valuable than coding oneself. The PM/spec writer becomes the new seniority.
3. The dev budget becomes pure OPEX. No more CAPEX on salaries: variable expenses, stoppable at any time, scalable by the hour.

The Nuance

Carson is a special case. Five startups behind him, a spec-writing capital accumulated over 20 years, a massive review discipline. A first-time founder won't replicate these results with simple access to Devin. The real leverage comes from methodological preparation, not the model.

And $20,000/month is a pre-product investment. It must pay off quickly — otherwise, it's unsustainable burn for a solo.

Key Takeaway

Carson's thesis: "Software engineering becomes a layer of agent coordination, not code writing." If this proves true on a large scale, it's a transformation as profound as the shift from hardware to SaaS twenty years ago.

To watch in the next 12 months: how many other solo founders will publish similar post-mortems? When a "one human + 15 agents" pattern spreads among indie hackers, the software industry has a real issue — or a real opportunity, depending on which side you're on.