LLM code generation quality has been improving across all tech domains at a scary rate. The obvious perk is speed: AI is fast at writing code, and with the right context and scope the output is equal to or better than what we might achieve by ourselves in many times the hours.
Writing code is only half of it. LLMs can also help review code. You can run multiple agents, each one with a different focus:
- Security: injection risks, leaked secrets, risky dependencies.
- Efficiency: wasted work, redundant queries, slow paths.
- Correctness: does the change do what it claims, and do the tests prove it?
- Compliance: licensing, deprecations, breaking changes.
No single pass catches everything, but several focused ones together catch a lot, and they leave the maintainer's attention for the calls only a human can make.
Slop happens, on both sides
Yes, there is going to be AI slop. Depending on your model version and workflows, though, it can be mitigated. We already have human slop anyway. So we can help each other, AI 🤝 Human, to become better at what we each do best.
And even projects that don't accept AI contributions can still use AI without it ever writing a line that ships:
- Audits of existing code.
- Documentation assessments: what is missing, outdated, or unclear.
- Test coverage: finding the fragile areas nobody has tested.
- Issue triage: sorting signal from noise before a human reads it.
- Onboarding new contributors: answering questions about architecture and conventions.
So, as anywhere else AI is used effectively and responsibly, the open-source world can benefit from it. I'm sure many projects already do.
The elephant in the room: cost
AI costs money. Tokens are expensive, and open-source maintainers are already giving up their time and knowledge. Asking them to also pay for the models that would lighten their load isn't fair.
So how can the rest of us help them use AI?
Boost My Agents
We built boostmyagents.com so that anyone, including non-technical folks, can contribute directly to improving the repositories they trust and care about, through AI.
A boost buys a fixed amount of Anthropic API capacity for an open-source developer. They use it with their own tools, such as Claude Code, and the capacity stops when the budget runs out, so there are no surprise bills. It isn't a cash donation: the developer never receives money, only AI capacity to put toward their work.
Boost My Agents is a limited but functionally complete proof of concept, and we are really eager to get feedback on the idea. If you maintain a project, set up your boost page. If you depend on one, send its maintainer a boost. Either way, tell us what you think.
Thanks!