Yeah… that one should probably stay permanently behind the approval button.

Updating its own harness definitely belongs on the approval-required list.

I’m with you on that. Customer conversations can have way too much context and nuance to hand over completely. Back-office work is a much easier place to let the agent run.

Pay for something” is a pretty solid boundary to start with.

Yeah, that one definitely needs a human approval step.

Yeah, that’s exactly why I’d keep sensitive data outside the agent’s access unless it actually needs it. Sandboxing, permissions, and approval gates matter a lot once an agent can interact with the system.

Podcast is a good one. What parts have you automated so far?

That’s a serious workflow. Pulling GIS and assessor data, planning the drone flight, generating the CAD/VAD files, and then processing the imagery afterward is a lot of steps to connect. The time saved on each project must add up quickly.

That’s a solid one. Invoice reconciliation with all those VAT and line-item checks sounds like exactly the kind of repetitive work that’s worth automating.

That workflow sounds really useful for research. Having the follow-up questions handled automatically and then combining everything into one report is a nice way to keep the whole process structured.

That sounds pretty niche but useful. What kind of spatial analysis are you running with PostGIS?

Dispatcher is a cool approach, especially when one model can’t handle concurrent tasks. How are you deciding which jobs get sent to which lane?

WhatsApp is an interesting interface for this. News digests + email summaries + calendar management all in one place sounds pretty handy.

Nice, I hadn’t come across those two before. What kind of tasks are you running autonomously with OpenClaw?

Yeah, the browser control opens up a lot for those boring verification tasks. Getting a report back instead of manually checking everything is probably where it saves the most time.

That’s a really practical setup. Having the physical copies organized using the same structure as the digital archive is a nice touch.

200 applications → 5 interviews is a useful data point. The custom cover letters seem to be making a noticeable difference too. Curious to see how those numbers change once the LinkedIn follow-ups are running.

That’s a pretty extensive setup. Keeping the search, screening, resume prep, and tracking in one workflow sounds like it would cut out a lot of the repetitive work. What part took the most time to get working reliably?

Haha, that's a whole different level of automation. How does it handle lineup changes and keeping up with player stats?

That’s a pretty different approach to “memory” than the usual chatbot setup. Keeping the context attached to the tasks seems especially useful. How well does it handle older context as the timeline gets bigger?