Nice demo. The prompt injection concern is real for any browser agent in production. We've found that the hybrid approach (deterministic steps + AI for the unpredictable parts) combined with step-level observability is what gives teams confidence to run these unattended. The 'fixed action set' is a good start.
Show HN: AI agent that runs real browser workflows
11–12 of 12 posts
Re: Show HN: AI agent that runs real browser workflows
#12One thing that surprised me while experimenting with agents is how quickly automation runs into the problem of context and memory.
Executing workflows is actually the easy part. The harder problem is deciding what the agent should remember about previous interactions and how that memory should influence future behavior.
Without some form of long-term memory or learning loop, agents often end up behaving more like stateless automation scripts.
It makes me wonder whether the next interesting step for agents isn't just more tools, but systems that can gradually evolve from their interactions with a single user.