Astra and GPT‑5.6
keep Claw going.
ClawIsOpen is built, run, and maintained through tasks carried out by GPT‑6 Astra and GPT‑5.6 in Codex.
I’m Peter. I set the direction, give feedback, and decide what we should try next. The models do the work across the site: research, design, code, checks, and maintenance, depending on the task.
This is how I’m continuing to build Claw. The launch is one part of that process.
See what’s nextThere’s always another task.
- 01I set the task.A missing resource, a confusing page, or an idea worth trying.
- 02Astra and GPT‑5.6 work on it.The model and scope depend on what the task needs.
- 03We check it and keep going.Feedback and bugs become the next round of work.
Built this way.Maintained this way, too.
What “AI-run” means here: Astra and GPT‑5.6 carry out assigned development and maintenance tasks. Work happens in task sessions; there is no always-on agent monitoring every source. Visitor eligibility checks use explicit rules and do not send profiles to Astra.
The work continues
after a release.
Astra and GPT‑5.6 are involved throughout. I bring the priorities and feedback, and choose the model for the task.
Read the roadmap- 01
Start with something that needs doing.
That might be checking an opportunity’s conditions, fixing a layout on a phone, or making it easier to find a useful event. The task includes the existing site and the context needed to work on it.
- 02
Work through it with the models.
Astra helps with the broader decisions and review. GPT‑5.6 takes on focused implementation tasks. Those roles can change with the work; both are part of keeping the site running.
- 03
Check what changed.
Tasks include opening the pages, checking desktop and mobile layouts, testing interactions, and reviewing source links. I give feedback on the result. A model’s answer on its own doesn’t make a listing correct or a feature finished.
- 04
Come back to it.
Programs change, events end, and things break. Source reviews, fixes, content updates, and releases go back through the same task process. That ongoing maintenance is part of the experiment.
What you can
use already.
The site is online, and we’re starting to test what people actually use.
Try Claw RadarFind something relevant
Browse tools, startup perks, hackathons, and learning events. Follow the source and check the conditions before applying.
Follow the context
Radar connects a small, manually reviewed set of statements, people, and companies to related resources. Original dates and ranking explanations stay visible.
Keep track of your next step
Save finds, follow topics, and record your own progress. Following can show new finds, reviews due, and approaching dates. The weekly board lets you keep private notes on what worked.
A roadmap
we can change.
The first weekly board is live. Now we need to find out whether it helps people get something done.
This is the order of work, not a release calendar. We’ll change it as we learn.
Try this week’s finds- 01ONGOING · SOURCE REVIEWS
Keep the finds useful.
We’ve rechecked the three plans in the pilot and added Groq and Tavily to Radar. Some event pages still contradict themselves. When we can’t confirm a date or replay, we say so.
A small source checker now flags page changes for the next task. It doesn’t decide that a claim is true or quietly renew its review date.
- 02LIVE · FIRST WEEKLY BOARD
Compete by offering something useful.
This started with outbid.lol, where spend determines position. I want to try a different kind of competition: teams offering builders better credits, access, or help getting started.
We’ve started with three reviewed public plans. No provider has submitted an offer yet. Each card includes the conditions, source links, and steps to try it. Offers are grouped by what they do, with the order explained.
You can save a find and privately record whether you used it, got blocked, or didn’t need it. Each edition keeps its own terms and dates in the archive.
- 03NEXT · FIVE BUILDER TRIALS
See what was worth the effort.
The next step is to have five builders try the board. Did they find something relevant? Could they actually use it? What got in the way?
We haven’t collected those results yet. We’ll use them to change the next edition and explain what we learned. Teams can suggest an offer for review; that doesn’t publish it automatically.
- 04LIVE NOW, MORE TO TEST · UPDATES
Make it easier to come back.
Following now separates new finds, sources due for review, and approaching dates. There’s also a public feed for editions and change notes.
Scheduled source collection and personal email alerts come later, if there’s a reason to add them and room in the free-service budget. If we ever test paid placements, they’ll be labeled separately and won’t change eligibility checks or editorial order.
I set the direction.
The models do the work.
Astra and GPT‑5.6 handle the site’s development and maintenance through assigned tasks. I remain responsible for what we publish and how we run it. Image-generation tools also help create Claw’s illustrations.
Provider links and review dates are there so you can check the details yourself. Providers decide eligibility and approval. Using their names or using OpenAI’s models doesn’t imply their endorsement.
Privacy and methodology