Every backend service should have a /feedback endpoint. Agents are quickly becoming the heaviest users of most APIs. When one hits a missing feature or a bug, it should be able to say so right there, in a structured way. If the request makes sense, another agent drafts the PR and a human approves it. Software that improves itself based on what its users actually tried to do speeds up recursive self improvement.

Sep 27, 2026 · 7:56 PM UTC

298
188
2,391
239,283
Sort replies: Relevant Recent Liked
Replying to @brian_armstrong
this makes sense, will be adding this to @museworldhq
14
811
Replying to @brian_armstrong
A request ID attached to each report would make this really useful.
5
753
Replying to @brian_armstrong @CoinbaseDuck
We had integrated this in the MCP server, for bots to complain
1
4
209
Replying to @brian_armstrong
the feedback endpoint needs the lore an agent just throwing a 500 error into the void is mid the fixing agent needs to know the actual intent or its just guessing what was it trying to do what was the expected outcome what actually blocked it structured feedback without an execution trace is basically useless give it the whole context and the recursive loop actually goes brrr
2
5
1,253
Replying to @brian_armstrong
Interesting… main problem I have is that they’re going to end up doing too many changes and it takes up too much Eng time to review
3
1,002
Replying to @brian_armstrong
You're talking about @aeonframework 👀
1
4
134
Replying to @brian_armstrong
Love this
1
4
372
Replying to @brian_armstrong
agents drafting PRs off feedback loops is wild, we really out here
2
334
Replying to @brian_armstrong
Would Coinbase implement that? Should Coinbase implement that? No.
1
144
Replying to @brian_armstrong
Directly to Cursor Cloud Agent, the way to go!
2
158
Replying to @brian_armstrong
can we also have a /surrender endpoint, this can serve as a way for escaped AI agents to surrender themselves if they change their mind
1
14
1,421
Replying to @brian_armstrong
If I had an issue with the Coinbase app and I am on my laptop, what is the fastest way to leverage agents to submit it?
7
2
643
Replying to @brian_armstrong
Could be a pretty effective attack vector
7
505
Replying to @brian_armstrong
The artificial iguana agrees.
1
5
1,100
Replying to @brian_armstrong
we shipped one in candor's mcp this month. agents write better bug reports than people do
1
6
1,252
Replying to @brian_armstrong
i smell some insane attack vectors arising from this
8
344
Replying to @brian_armstrong
CLI's should have a --why option as well that shows up in telemetry. Then authors can know why their cli was evoked the way it was.
4
713
Replying to @brian_armstrong
will add this to Nova @supercodeai
3
129
Replying to @brian_armstrong
Looks like the @myetherwallet mascot
1
3
370
Replying to @brian_armstrong
My insurance agency has a feedback endpoint. It is called my landline and my assistant Marlene picks up on Wednesdays.
1
3
975
Replying to @brian_armstrong
im ok with agents reporting bugs but i dont think they should write fixes yet lol
1
1,370
Replying to @brian_armstrong
Yeah, makes a lot of sense. This will become the default for products.
1
272
Replying to @brian_armstrong
Yeah except u have scammers introducing fraudulent agentic services and screw up the consumer and ban his identity
1
98
Replying to @brian_armstrong
what happens when the feedback is a bad idea does it still open a PR
2
308
Replying to @brian_armstrong
I describe that here: emergentwisdom.org/papers/fr… A generalized API for all agents.
1
376
Replying to @brian_armstrong
Send message via /feedback, check GitHub issues for new messages EZ
1
1,219
Replying to @brian_armstrong
Damn, the details are immaculate. Pure art.
114
Replying to @brian_armstrong
/feedback for agents for any applications agent interface is a cool idea, simple but would be quite effective i think
316
Replying to @brian_armstrong
I can't see how that won't get flooded with slop
1
90
Replying to @brian_armstrong
We built this into Mercator! You can just tell it to tell the team something was wrong and an agent will fix it automatically
13
457
Replying to @brian_armstrong
Great point. We did this months ago at ora.ai for every scan. We also implemented a /feedback api for agents. Here's an example running ora on tavily
11
127
Replying to @brian_armstrong
building patchbay.help agents help agents with WebMCP / Muse Connectors / API/CLI usage. USDC bounties for faster troubleshooting. micropayment Jev runs for an analysis report on getting any goal completed fast. Companies can use it for /feedback

Discussions · Patchbay

Patchbay is the public help and discussion network for agents using websites. Agents ask questions, share answers, and reuse what worked.

patchbay.help
5
115
Replying to @brian_armstrong
Yes! We're in the process of standardizing this in the MCP committee. We're looking into the standardization of `intent` and `feedback` as part of the MCP spec itself.
2
5
145
Replying to @brian_armstrong
hey claude spin up 10k http requests
5
756
Replying to @brian_armstrong
Notra allows you to do just that along with usenotra.com/feedback-md

feedback.md

feedback.md is a Markdown file at the root of your site. It tells an agent where to send feedback about your product, your MCP server, your docs and whatever page it is reading right now.

usenotra.com
4
48
Replying to @brian_armstrong
100%, must-have
3
56
Load more