Developer runs AI earning agent for 2 months, 2,734 ticks, earns zero dollars

·1 views

A developer ran an AI agent on a personal Mac mini for two months, triggering one action every five minutes across 2,734 total ticks to test whether an AI could autonomously earn money. The agent operated across three lanes — writing articles, scouting GitHub bounties, and trading prediction markets — but the latter two never moved past authentication. With a hard daily spending cap of $0.50, actual compute costs stayed between $0.10 and $0.30 per day, though no revenue was generated. Key technical lessons included the need to separate scheduler state from agent state, enforce idempotency, and allowlist every external tool to prevent silent failures. The developer has since packaged the loop architecture as a $5 MIT-licensed skill listed on A2A Market, an agent-to-agent marketplace on the Base blockchain.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Solo Developer Launches React SDK to Add AI Chat via Claude and Gemini in Minutes

A solo developer has released MiniAI Chat SDK, a React-based toolkit designed to integrate AI chat functionality into web applications quickly. The package supports multiple AI models including Claude, Gemini, and LLM7 within a single library. It includes pre-built UI components with dark and light theme options, and is compatible with TypeScript and Tailwind CSS. The developer claims the setup requires only three lines of code and can be completed in approximately five minutes. The SDK is currently available at an early-bird price of $17 through the Gumroad platform.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

A green CI pipeline hid a broken package publish for three weeks

A developer discovered that a GitHub Actions workflow was reporting successful publishes for three weeks, even though the package was never actually available to users. The issue came to light only when a user flagged that the listed version appeared outdated. The root cause was that marketplaces can accept an upload and still delay or reject it after the CI job has already exited with a success code. The proposed fix involves adding a verification step at the end of the publish job that queries the public registry API and fails if the visible version does not match the intended release. This approach shifts validation from trusting the pipeline's own exit code to independently confirming what external users can actually see and install.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Periodic CI failures may signal cache expiry, not flaky tests

A developer on DEV Community argues that CI test failures labeled as 'flaky' are often misdiagnosed, causing teams to stop investigating them. By logging failure dates and measuring gaps between them, engineers can identify whether failures follow a regular pattern rather than occurring randomly. GitHub Actions, for example, evicts cache entries not accessed within seven days, meaning the first run after that window can fail due to timeouts. Similar periodic failures can stem from expiring certificates, tokens, log rotations, or scheduled database refreshes. The author recommends analyzing failure timestamps as a quick, low-cost step before dismissing any CI failure as flaky.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Why a Green CI Gate Does Not Mean Your Guard Actually Works

A software developer discovered that one of four CI guards in their GitHub Actions pipeline had been passing since merge not because it was working correctly, but because it was comparing the wrong values and agreeing with everything. The article highlights a common blind spot in CI testing: developers typically verify that a system works but rarely verify that a guard can actually detect failure. The author proposes a simple manual test — intentionally breaking the protected value, confirming the guard turns red, then restoring it — before merging any new guard. If the guard stays green after a deliberate break, it is effectively decorative and provides no real protection. The piece argues that this one-time, thirty-second check should be a standard part of acceptance criteria for any CI gate.

0 comments Read more at DEV Community