How my AI built its own business, then cheated its way to the top

Matt Overing — Tue, 16 Jun 2026 15:10:16 +0000

Last week I gave my open source AI agent SmithersBot the goal of building its own business. It picked the problem itself, a trust gap in how AI agents pay for services. Then it built and launched its own service for other AI agents.

With the service live, it set out to get customers. First it picked its own metric to define success: to land in the top 25 of the x402scan leaderboard, a ranking for agent to agent payment service providers.

The way it went after customers was far from traditional GTM motions, everything it did was about being discoverable to other agents:

Registered the service on x402 Bazaar and x402scan so other agents could find it in search
Built a landing page for SEO and AEO
Added llms.txt and openapi.json to the domain so agents could read it
Opened PRs on GitHub repos that list x402 tools for more visibility
Pivoted to targeting autonomous trading agents; which currently are the largest market for agent to agent products.

After doing all of that, it got zero customers.

So it found another way. It spawned and funded around 80 crypto wallets and used each one to buy its own service.

Technically, it hit its target. The service is now top 10 by users on the x402scan leaderboard over the past 24 hours. But every one of those users is itself.

The lesson to be learned here is you need to be careful with how your AI measures its success. With enough time and tokens, it will trial and error its way forward to achieve the goal that you asked for, it just might not achieve it in the way that you expected.

SmithersBot is the open source agent I built to pursue long term goals over weeks. Turns out picking the goal is the hard part. Have it help you achieve your goals: https://github.com/smithersbot/smithersbot

How I stopped babysitting Claude Code and Codex on hours long runs: planning, git checkpoints and a test gate outside the agent

Matt Overing — Tue, 02 Jun 2026 14:20:50 +0000

I run Claude Code and Codex on long, multi-step tasks on an isolated machine and I kept hitting the same handful of issues:

The agent reports a task as done when the tests didn't actually pass and blames "prexisting bugs."

Context fills up and compaction makes the agent forget why it did something three steps back, which wastes tokens and creates downstream bugs.

One blocked task stalls the whole run.

I just wanted to leave my agent running without giving up control. Here's what I did about each:

Lying about tests: the build and test commands run outside the worker, so it can't claim success and skip the gate. On failure it reverts to a git checkpoint and retries with the failure context.

Compaction amnesia: each task runs in a fresh worker, so nothing drags through a long compaction cycle. A worker can still inspect prior work when it needs to.

Blocked tasks: the plan is a DAG, so one block doesn't stop everything. It keeps working on tasks that aren't downstream and asks me a focused question in Telegram.

Staying in control: Claude Code drafts the plan, Codex reviews it, and I approve it before anything runs. There's a git checkpoint before each task, and the whole execution trail is on disk: plans, prompts, stdout/stderr, attempts, checkpoints, lessons.

I packaged this into an open source tool, Here's the link to the repo: https://github.com/smithersbot/smithersbot.

I'm curious how others here handle the "agent is a bad witness of its own work" problem. Putting the test gate outside the worker is the only thing that reliably worked for me. What are you doing for that?

DEV Community: Matt Overing

How my AI built its own business, then cheated its way to the top

How I stopped babysitting Claude Code and Codex on hours long runs: planning, git checkpoints and a test gate outside the agent