OpenClaw scared me a bit and struck me as an “agentic is here” moment. Had I made predictions on what the last 7+ months would look like, there would have been a lot more obvious steps/updates by now. That said, I count myself as one without the background knowledge to accurately evaluate the technical advancements in AI. My reaction to OpenClaw might be just one way I’m wrong...maybe I overreacted.
I would love to hear from those better informed and equipped to address this. Was OpenClaw a big deal to you? Are you surprised that the last 7 months haven’t had more headline-worthy OpenClaw like moments? Or is this just how progress goes, in fits and starts? [Don’t throw tomatoes at me, but] could it even suggest AI progress might actually be slowing? Or that the low hanging fruit at least has been picked when it comes to LLMs and LLM-wrapped agents?
I would say it updated me towards longer time horizons. I expected AutoGPT to be a big deal. Instead, it didn’t really work that well. I expected OpenClaw to be a big deal, but it still isn’t that big of a deal. I live in Mountain View and talk to rationalists and programmers all the time, and I don’t know a single person who is having an AI agent do a substantial amount of their everyday computer errands outside of work.
I figured GPT 4 was smart enough to, like, be people’s personal secretary and randomly pop up with useful ideas for them, but there’s clearly something that is harder than I think about it.
It was a coalescing of things, not a direct update for me. I was deep enough in daily use for coding and infosec agents that much of the hype seemed like “yeah, this is 6-12 months early, and a horrible idea to trust too soon”.
I didn’t fully realize just how much of the moltbook and true hype was fantasy/grift (a few 10k people claiming to be 100s of K of autonomous bots making commerce and finance magic, at least in part to try to monetize random tokens like $CLAW and such). But I was skeptical that there was much there on that level.
It was definitely proof that far more automation and longer-running semi-supervised activities were on a shorter horizon than many of us thought, even in the business.
The last 7 months have been within my bounds for innovation—so much of the promise and hype have become more real. Actually functional autonomy has increased from minutes to tens of minutes due to longer context windows, WAY better harnesses, a lot of evolution of knowledge in how to build/monitor guardrails (skills, triggers, adversarial agent roles, etc.). And yes, because of the hype bubble of OpenClaw lighting a fire under the more responsible mass of developers and providers. And users.
OpenClaw itself didn’t, but the agentic systems have. Last year, I thought I’d have a robot assistant by now, and I’d be a software engineering company on my own by next year. Now, I regularly have a team of robots write my code.
I’m not a lot more productive because my coworkers haven’t really kept up, but I spend a hours less time sitting at my desk in my home office. I generally will spend about 10-20 minutes setting up the task, and then have them work on it, and I have them send me an audible alert when they are done writing and reviewing the work, at which point I’ll come back and review. Sometimes I leave comments, and the robots fix the issues, and tell me once all of the tests pass.
I can only run about 3 concurrent jobs at once due to memory limits on my computer (we have like 20 services in our system, and those docker containers use up a good bit of memory. Cursor too). I feel like i could manage about 6 at a time before the context switching would become a bottleneck
Somewhere around February was the inflection point. Agentic systems went from situationally useful to just how I work. The world has changed, and most people don’t know it.
Now I’m using my time to learn the mathematics of AI with the help of ChatGPT and Codex.
The future is wild.