正在加载视频...
视频加载失败
The best engineering teams are using cloud agents to build, test, and verify work, end-to-end. Basis builds AI agents that cut 20-hour accounting tasks down to 30 minutes. From 20-day agent threads to an office full of microphones, here's how they work with Devin.
13 条评论

For startups interested in building with Devin, we're offering $65k in credits – in addition to dedicated support – for select teams. Apply here:

@trybasis Love learning from Basis. They are constantly living 6 months ahead

@trybasis 20 hours down to 30 minutes on accounting tasks matches what I see as a small business owner. Back office was where AI paid off first for me, way before creative work!

@trybasis Build and test get the attention, but verify is the part that decides whether a team can trust the output. If verification stays manual, the speed gain stops there.

@trybasis Long-running threads usually struggle with context drift and state bloat over time. Curious how they handle context compaction across a 20-day run.

@trybasis That sounds like a glorious amount of accounting time to get back. I can understand the enthusiasm. ❤️

@trybasis Long threads still need a verifiable close.

@trybasis a 20-day thread isn't one thread, it's a stack of compactions, and every compaction is the model deciding what its future self gets to know. whatever bit you on day 18 got cut on day 3. that's the gap i built memoryrouter for.

@trybasis How are they isolating deterministic reconciliation logic from model hallucination when shrinking those 20-hour accounting runs down to 30 minutes?

@trybasis Let us use our Claude Code subscription and I will switch to it 👀 !

@trybasis The best engineering teams are using Devin, it's all they need.

@trybasis Can you have a sub tier with more opus usage, that would really fix everything. I am now running max plan, always stuck with usage limitation and have to topup more before reset. it can never go to full automation with this shit. I also running it with a chatgpt sub already.

@trybasis I run Codex every day, and drafting the change is the easy part. The test and verify loop is what I actually want more of. An agent that can ship a patch is useful. One that can prove it held is the whole job.
