A real game changer: gog CLI
Not only for my Claws, but for my local work as well.
Today I optimised Google Tag Manager, Search Console, GA4 and Google Ads in a single session.
Thanks @steipete
https://t.co/hKST0wHB20
I'm excited to see where @t3codes is going to take us.
The way they set up their CONTRIBUTING(.)md with the hoops to jump through and vouch:* is nice.
Scanning through the vouch:trusted PRs I can see where they are headed:
- remote access and pairing
- broader multi-agent support
- performance is a big deal for them
- mobile is clearly a direction, though not near term (most excited about this)
Bonus: the small things matter. Being able to see which session has an active terminal running, or seeing a PR status at a glance is really handy
I'm struggling to believe production level multi hour/day/week autonomous runs.
I got my longest single-run from GPT-5.5 in Codex App today.
It ran for 50min and stacked 4 PRs.
Trying a new method for prompting and it looks promising:
1. Visit the {PLATFORM} API docs at platform(.)com
2. Before writing any code, list the testable behaviors for this feature. For each one, describe the test in one sentence: what's the input, what's the expected outcome. Don't write code yet, just the list.
3. Then write those tests. They should all fail initially since the feature doesn't exist yet.
4. Then implement the feature and make them pass. Don't weaken any assertion to make a test pass
Repeat for each milestone
T3Code: @theo this is almost my favourite app right now
1. Activate Local Backend Access
2. Serve through Tailscale/Cloudflare Tunnel
3. Open in Safari iOS, save bookmark as Web App to Home Screen
Mobile still super clunky, but its working and the potential is 😮💨
- The same session on mobile or computer (not either/or like Happy). Continue wherever
- Codex + Claude in the same UI
- Can even add a remote T3 server
My weekend project with Codex xHigh:
Step 1: Force real test coverage
Step 2: Use mutation testing to validate the tests
Step 3: Drive down complexity with a CRAP-style metric
Seeing my wife create her own Lego collection website with Codex.
Working and refining to add new features - loving every second of it
Never thought I’d see this beautiful day 🥹
@JosvdWest@karpathy For me Obsidian solves the eyesore of reading and editing raw markdown files. That UX is bad in an IDE, so Obsidian gets you close to Notion-like visual editor while still have access to the raw markdown
One new thing as a solo dev is finding new things in my codebase.
(100%) Pre 2022: I knew my codebase like the back of my hand
(100%) 2023: Still knew but had a guest in the form of command and tab
(90%) 2024: More new code, but still in control since I was using ChatGPT Project and copy pasting
(50%) 2025: Moved to Agents, started losing control
(30%) 2026: Moved back to Projects, but copied and pasted the full output to a worker model, at 30% control
(est 0%)End 2026: I think if Jules from Google gets it together, I will always be in a state of new code in the code base that I had zero involvement
I don't even look at myself as a solo dev, my company is just my agents and me.
I'll make more before I hire someone, not that I'm making enough to hire someone.
My code understanding is moving back to being part of a team and finding new bits to the code after coming from holiday break.
@RyanClogg David Ogilvy - Confessions of an Advertising Man
Jason Flatlien - One to many: The secret to webinar success
Russel Brunson - Expert Secrets
Robert Cialdini - Presuasion
Donald Miller - Building a Storybrand