AutoClaw
An open-source framework for deploying and operating AI agent systems.
Recorded experiment outcomes
The framework hub, playbooks, first article, and a deployment were recorded as shipped. The second HN publication, Agent minimalism, shipped on 2026-06-22 (publication tracker). Upwork validation last recorded 71 bids, 3 conversations, and no contracts; its final ROI and continue/adjust/exit decision remain unrecorded. Track experiment follow-through in leadlinks #37; summarize the traction decision here rather than maintaining v0.2 and Upwork project pages.
Historical planning — March 2026
From March 2026 planning Anu & Rufus
Key docs:
Summary: Immediate Execution Priorities
- Deploy with real users immediately
- Internal + close network (e.g. partner, team members)
- Goal: working instances, real usage, fast feedback
- Acquire early external clients (e.g. Upwork): Small number, used as forcing function, not primary goal
- Ship content + distribution quickly
- Record raw tutorials within days (not polished)
- Publish alongside:
- Website (topics + tutorials)
- Git repo (code, examples)
- Create traction-oriented assets
- e.g. “awesome deployment” repo
- Designed for visibility on Hacker News
Operating Principle
- Speed over polish
- Working system + public artefacts within a week
- Combine:
- Real deployments
- Public learning (content, repos)
- Lightweight client work
Next steps for getting started
- Finalize 2 PRs: immediate one on pivot (draft below) and PR2 imagining this has worked in 2 months
- Anu outflow about what it takes to deploy right now from his experience (including the pain points we encountered with autopilot)
- This will generate a lot of items we could write posts / youtube videos about
- Agents SRE is the future
- Note note just about openclaw
- Anytime you deploy an “agent” e.g. a chat interface (e.g. queryless) you want this …
- E.g. why kiloclaw vs deploy yourself or on your own org’s infra
- Where you can deploy yourself (what is needed, what are the options)
- …
- Agents SRE is the future
- This will generate a list of things to research: that we don’t know the answer to …
- This will generate a lot of items we could write posts / youtube videos about
- Focus on working with real users immediately and that will tell us what is actually important to them … (vs what we think)
- Have Luccas get something live in next 24h for Daniela or Osahon …
- Joao work on Upwork (has days to week lead time)
- => Anu: drafts at least the first Press Release (preferably both)
Second set of next steps (at top of draft SCQH)
From Anu's draft SCQH https://docs.google.com/document/d/1KTYJaB7zalmZ_YZLGEXnBZxfaWvwOD7ezVfZM5x0kPc/edit?tab=t.9nqhm7sqa5d2
- Revise the SCQH
- [ ] Clarify north star and metrics for that …
- [ ] Time estimate:
- [ ] Aside: Hypothesis should ultimately be one short paragraph which says something like e.g.: Using 2 person weeks and by the end of April we can XYZ …
- [ ] Review from Rufus (?) - Write the 2 PRs
- [ ] HN announcement in one week:
- [ ] Pivot / learnings
- [ ] Here’s a github repo …
- [ ] Pattern / repo for deploying
- [ ] Patterns for particular tasks
- [ ] Maybe a mix of handbook and skills
- [ ] How to back it up
- [ ] How to secure it …
- [ ] How to give it memory
- [ ] How to bring your own key and/or track usage …
- [ ] How to add to telegram (groups)
- [ ] What is it: more like handbook / website / skills / some scripts
- [ ] Aspiration: could be a framework …
- [ ] Recipes / playbook
- [ ] AutoClaw is an open source playbook for deploying and managing agents like OpenClaw
- [ ] We are not about the agent …
- [ ] We are about how to deploy them
First press release: OpenSource Playbook for openclaw (and agent) Deployment
What we discovered building AaaS prototype in the last month and a half and what we are doing now
When Open Claw came out, we started using it immediately and began thinking about deploying it. We were using it not only for ourselves, but also as a basis for building products like this. It was a kit for deploying agents.
Now, when you use Open Claw and then use something else, this is coming up. What we actually built was our own little SaaS for this. Here you can see a video, or here's what we built. We launched it at the beginning of February and tried it out with our team and some other potential users. [TODO: insert video]
What we saw were certain pain points, for example:
- We have to manage AI billing and get our own subscription to AI, then fold that into users so they're paying twice (they've got their own AI subscription).
- They have to bring their own key, but there are complexities around the API.
In summary lots of complexity … and we think a framework is better
There is a load of complexity that we encountered when we actually tried it out. What we have come to conclude is that it's not impossible to do SaaS, but what we really want is a framework. We want patterns and a playbook for deploying AI, because you are going to want to deploy it for your own organisation. You don't want to use a SaaS platform; you want this inside your organisation, talking to your own data and content. That's what makes the agent useful.
There are all of these things. Our view at the moment is that it is more about something open source: a playbook and tutorials that are about patterns. We are pivoting, or not pivoting; we are just moving forward by open sourcing all of our learning and sharing it, and inviting others to share their playbook. How do they do agent deployment, like Open Claw deployment?
What are all the sharp edges? For example …
- You need to back up your agent.
- You need to give it memory.
- You need to sandbox and give it access to certain tooling.
And overall how do you just keep up with what is happening …
You can even say that, in the last month, we've seen this announced, this announced, and this announced, but just keeping up with what's going on is like a full-time job. Nvidia releases something, these people release something.
Anyway, what we're hoping to do with this project is to share our learning as we go along and create an open source playbook and maybe a framework that people can use to deploy agents for themselves or for their organisations at some kind of scale, with a lot of good defaults (sane defaults) for how to do things.
Tidy Transcript for priorities
So, I would be like, straight away, we want a client internally — or like Sylvie, my partner — but someone we’re actually going to deploy one for.
Then I would try to go to Upwork and recruit a few clients.
And thirdly, I would try and launch — I would brainstorm a list of tutorial YouTube tutorials that we would do. We’ve already got some good ones, like the GitClaw one.
And then it could be super raw. We are going to record — in the next three days — just a first pass. We write a script and record them. They’re not polished. It’s literally like someone interviews you — you go on Riverside — and you just interview each other. You share your screen and go through it. Riverside will auto-edit to some extent.
I would just do that. Get some content out.
Then I would put up a website with the tutorials and the topics that we think matter.
And a Git repo with examples or skills or something like that.
Another hack would be something like an “awesome claw deployment” Git repo. I’m sure there’s an “awesome OpenClaw” already, but something like “awesome claw deployment” may not exist. The point is: things that get traction on Hacker News — reference projects.
I’ve had success with that before. For example, with Web3 I did “awesome Web3 critique” — not “awesome Web3” — and that got about 2000 stars.
So I’m just brainstorming, but the point is: things that generate traction.
And I would say: by the end of the week, we have deployments running with real people. Daniela has one, Sylvie has one, your wife has one, Monika has one — we just have something working.
That’s what I would push for.
Log
March 2026
March 2026 Google doc where Rufus and Anu got an outline: https://docs.google.com/document/d/1KTYJaB7zalmZ_YZLGEXnBZxfaWvwOD7ezVfZM5x0kPc/edit
This was originally Anu SCQH draft with a bunch of
Recorded work and operating context — September 2026 scope review
- AutoClaw v0.1: Historical February–March version, recorded done. The page had no substantive body. Historical source.
