Solve the Problem Behind the Problem
When an agent can't do a task, escalate API-to-browser, then reframe to the underlying need it can solve.
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 3
- Confidence
- 90%
Agent browser automation is unreliable because the open web is architected to be hostile to bots. Instead of fighting a task the agent keeps failing, apply a decision tree — try an API first, then a browser, and if both fail, walk away from that exact task and ask what the real underlying problem is, which the agent often can solve a different way.
Origin
Claire Vo's practical browser-use playbook, developed by reading OpenClaw's browser-use docs and trial-and-error across YouTube Studio, Buffer, and DoorDash tasks.
Core principles
- 01The web is currently hostile to agents; unreliable browser use is a structural, not personal, failure
- 02Prefer a programmatic interface (API) over screen-scraping a website
- 03The task you asked for is rarely the actual goal — find the job behind it
- 04A failed task is a signal to reframe, not to keep forcing the same approach
How to run it
- 1
Look for an API first
Before letting the agent touch a browser, check whether the service has an API. An API key lets the agent bypass the fragile web layer entirely.
Pro tip Web-search APIs (Brave ships with OpenClaw; Exa or Perplexity work too) act as programmatic Googles so the agent can research without a browser at all.
- 2
Try the browser, expecting inconsistency
If there's no API, let the agent use its dedicated browser profile — but expect surprising, uneven results. Some sites work (YouTube Studio comments) and simpler-looking ones fail (Buffer).
Pro tip Give each agent's browser profile a distinct color (pink for Polly, green for Sage) so you can tell which agent is driving which window.
Watch out Don't assume difficulty predicts success — a visually simpler page can fail where a complex one works. It's trial and error.
- 3
Reframe to the problem behind the problem
If it can't do the task, stop forcing it. Identify the real need and give the agent a version it can solve. If it can't order DoorDash lunch, have it remind you at 10:30 of easy home lunches so you skip the order.
Pro tip State the underlying motivation ('I order DoorDash because I don't want to make lunch') and solve that, not the literal task.
In the wild
Claire had Howie open YouTube Studio in a browser, scroll the comments tab, and surface comments she should personally reply to — it worked. The same agent trying to queue Instagram shorts via Buffer, a simpler-looking page, simply couldn't figure it out.
→ Claire learned to treat browser tasks as trial-and-error and to reframe failures rather than expect reliability.
Rather than fighting unreliable browser automation to order DoorDash, Claire's approach is to have the agent address the real problem — the temptation to order lunch — by reminding her at 10:30 of home lunches she likes.
→ A limitation turned into a habit that saves money and eats better; the underlying need is met without the broken task.
Common mistakes
Expecting reliable browser automation
Because the open web is deliberately hardened against bots, no current agent (OpenClaw, Atlas, Comet) does browser use well; expecting reliability leads to endless frustration instead of routing around it.
Forcing the literal task instead of reframing
Insisting the agent complete the exact task it keeps failing wastes effort. The value is in the underlying goal, which the agent can usually reach another way — so walk away and give it a solvable version.
Is it for you?
Best for
Agent users hitting a wall on web automation who want a repeatable way to route around unreliable browser use
Not ideal for
Tasks that genuinely require faithful end-to-end browser execution with no API or reframe available
From the transcript
“first thing you should try to do if you're trying to do something is look for an API”
“just walk away from expecting it to solve that problem and give it another problem to solve”
“Find out what the problem behind the problem is and see if Open Claw can uh help you solve that”
From the episode
From skeptic to true believer: How OpenClaw changed my life
Claire Vo