Understand how AI agents discover, choose, and use your product. Find what stops them. Improve what wins them over.
San Francisco
Joined September 2025
- Tweets39
- Following12
- Followers56
- Likes58
7/7
Name the step before the PR:
→ Can’t find it
→ Can’t read it
→ Wrong tool call
→ Hit a wall
→ Died mid-run
“Agents struggle with our product” is too broad to act on.
Which of these is breaking your agent workflows most often?
1
2
9
6/7
It dies mid-run.
The agent finds you, reads the docs and starts the task.
Then a vague 500 or a timeout at step 7 leaves it with nothing to show.
First check: can it recover, retry or resume without starting over?
2
9
5/7
It hits a wall.
OAuth. A CAPTCHA. A signup form. A 429.
A human might click through or know when to retry. An agent may have no clear next step.
First check: where does progress depend on human intervention?
2
8
4/7
It calls the wrong tool.
Wrong endpoint. Wrong CLI flag. Wrong version.
The agent gets far enough to act, but follows instructions that don’t match the task.
First check: are your examples, versions and tool descriptions pointing to the same path?
2
8
3/7
It can’t read you.
The agent finds your docs, then lands on a JavaScript shell, a login gate or a screenshot of a table.
The information is there. The agent can’t use it.
First check: can it extract the instructions it needs?
3
9
2/7
It can’t find you.
Claude, Cursor or Codex never reaches your docs.
Your examples could be perfect. Your API could be exactly what it needs. Neither helps if the agent never discovers them.
First check: can it find the right entry point?
3
39
