OpenAI built a new Git backend because AI agents sped up the work, and the bottleneck moved to testing, review and decisions.
Why this is useful to know
OpenAI built a new Git backend, Pando, because its coding agents pushed so much work through the development systems that the infrastructure struggled to keep up. According to OpenAI, median push latency dropped by nearly 80 percent. The author's main point: when code becomes cheap, the bottleneck moves. At FurtherAI, merged changes per engineer rose fivefold, but the waiting moved to testing, review and the question of what you should build at all. For a small or mid-sized business this means: more output does not automatically bring more value. Look at where work gets stuck, also outside IT.
When you need this
Say your team builds faster with AI, for example prototypes, automations or changes in Odoo. Then more work lands on the shelf that needs checking and deciding. What can you do? First: list your last ten deliveries and note for each how long it waited and where. Second: set fixed review rules that an agent can check by itself, so a person only looks at the real judgement call. Third: feel free to throw prototypes away when they deliver nothing. A working demo is a question, not a finished product. Measure lead time to value, not the number of changes delivered.
How it works under the hood
Agents fetch code, run tests and submit changes, often at the same time in one monorepo. That creates a lot of extra push and read traffic on the Git backend. According to the author, Pando rests on two design changes; OpenAI described them on 2 November in its own engineering write-up. For your own setup it means: the waiting then often sits in testing, review queues and decisions, not in the quality of the model. FurtherAI turned repeated review comments into 21 automated checks. Agents fix the errors those checks find before a person looks. In effect that is a guardrail layer around the agent. If you want to know the architecture of Pando, read OpenAI's own write-up.
This news touches chapter 23 and 25 and 14 of the course AI(l) or nothing by Newway. If you follow the course, this subject returns there at a calmer pace.
What you can do with this
Want to know what this means for your business? Plan an introduction: half an hour, no sales pitch, but an honest answer to the question where to begin. Rather read on first? The course AI(l) or nothing is free and starts with the basics.
Written in response to a publication by Nate's Newsletter (Substack): https://natesnewsletter.substack.com/p/openai-pando-git-backend
