John Potess

Traveling around and building stuff

[John Potess @johnpotess]

I can't believe I hadn't explored using DeepSeek sooner, switched over from gpt 5.4-mini to v4 flash for internationalization on DiveJourney, and with roughly 14k pages to translate the savings were substantial.

It took a few rounds of dialing in the prompts and evals, but in the end this is easily going to enable adding a quite a few more languages than I was initially planning at a fraction of the cost.

And it was super easy to try out diff models with @OpenRouter

[John Potess @johnpotess]

I ended up getting claude max for the month (about 5 hours before the Fable fiasco). But it's been interesting working with both codex and claude today.

So, not a total loss

[John Potess @johnpotess]

just waiting for a 5.6 release

[John Potess @johnpotess]

@thsottiaux @thsottiaux will this also be made available to Pro users? 🙏🙏

[John Potess @johnpotess]

I sat down to write about Linear, sprint management, and automation. Then I put on music, looked out over Bangkok, and wrote this instead.

New post: The Power of the Vibes
https://t.co/AiXOJmnmGo

[John Potess @johnpotess]

@OpenAI Would be great to have this for Pro users as well.

[John Potess @johnpotess]

@Yinielin @thsottiaux That sounds pretty interesting, where can I see more info?

[John Potess @johnpotess]

I feel like there is just so much to keep track of and so many moving parts now.

You can ship so much with AI, and building new features is almost trivial code-wise, but I’m running two products now and it’s getting wild trying to keep track of everything I’m supposed to be doing to move them forward.

I generally structure my life in a 12 Week Year setup and manage monthly, weekly, and daily burn-downs and goals in Notion.

I’ve had some success offloading update gathering, sprint organization, and outreach into an internal CRM + OpenClaw to help manage it. As a personal assistant for everything else, mixed results so far.

How are other people handling this?

I’m guessing there are tens of thousands of people trying to juggle the same build / learn / compete / keep up / manage life / stay sane problem.

Any suggestions on flows, tooling, or what’s actually worked for you?

[John Potess @johnpotess]

I've started to plan my week's codex token burn with the chance that there will be another usage reset 😅

[John Potess @johnpotess]

@stephenhaney Really awesome, I've been using it all morning with codex and just truly impressed by how smooth and fun it is to use it.

[John Potess @johnpotess]

this is amazing - I tried it out this morning and blew my mind - exactly what i didn't know I needed 🔥

It's magical watching it appear your site into Paper. I feel like this is going to unlock a ton for solo builders

[Stephen Haney @stephenhaney]

Hello!

Today we're releasing Paper Desktop

Paper is now a canvas for Cursor, Claude Code, Codex. Any agent can read and write html to Paper.

• push or pull from your codebase
• pull real data from anywhere
• less work, more design

What will you ship? Sound on 🎶

[John Potess @johnpotess]

Lots of nice DX updates in new codex cli (0.105), themes, and this little sub-agent selector with names / easier to manage sub agents 🙏

I'm most excited for the new voice transcription feature.

This is is basically the only thing i use voice to text apps for currently - so quite ideal if it's now a built-in part of the CLI.

[John Potess @johnpotess]

I've been working out systems and processes for my OpenClaw agent (Jerry) this week

so wild when you see it actually saving you time, I'm pretty excited to see the output improvements on my workflows, will write up soon

[John Potess @johnpotess]

the note ate the end for the 2024/2025 fit curve is wild

if that turns out to be true, 2026 is definitely the year to lock in hard as a builder to be ahead of the wave

[METR @METR_Evals]

We estimate that GPT-5.2 with `high` (not `xhigh`) reasoning effort has a 50%-time-horizon of around 6.6 hrs (95% CI of 3 hr 20 min to 17 hr 30 min) on our expanded suite of software tasks. This is the highest estimate for a time horizon measurement we have reported to date.

[John Potess @johnpotess]

@matt_elevenlabs 11B

[John Potess @johnpotess]

January Shiplog for DiveJourney is live 🚀

What shipped:
- Rebuilt dive map filters (country + destination)
- Faster map + destination pages
- Shareable trips, recaps, and achievements
- Anti-spam + email hardening
- Expanded destinations across Mexico, Thailand, Australia, and Brazil
- Stronger Featured Destination Partner flow for dive shops

Full focus on shop activation + distribution now.

https://t.co/hpbGzJte9Q

[John Potess @johnpotess]

Walking around Bangkok messed with my AI doomscrolling brain:

Maybe I’m in a bubble of what people think is coming and the access and exposure to AI tooling and the advancement in AI-related areas.

But man, it feels like we’re all scrambling for a little piece of this pie that’s about to expand and consume everything.

And that’s how I feel until I go walk around outside.

It's like how do you automate all the street food vendors and people selling small goods out of their little storefronts, r one-person nail and barber shops? How much time does it take to convince different cultures like this that these things should be automated, or even how they could be?

Maybe it’s easier in the US, especially in remote/online/tech industries. We’re max isolated there - so yeah, makes sense - if you don’t need to interact with anyone, and you don’t have to see someone in person, why wouldn’t you automate that out?

Maybe the happiest populations, at least until everything is consumed, will be the cultures with more interpersonal connections and interactions.

Anyway, glad i'm here.

[John Potess @johnpotess]

I used to listen to music all day while coding and working - basically constantly whenever not in meetings - in a more or less flow state.

In the last six months since working primarily with claude/codex I've noticed I basically never listen to music anymore.

Which is a bit of a bummer, but I think it's because of the amount of focus writing specs and reading outputs.

Anyone else experienced the same thing?

[John Potess @johnpotess]

Ah yeah I mean more so codex debugging it's own UI issues. Before I'd get weird UI choices that didn't make any sense or heavily drifted from my original spec on long runs. So this seemed to keep it aligned much better to have these check-ins running.

I'm looking at adding in an initial design step + lightweight design review during runs and see how locked-in I can get it.

[John Potess @johnpotess]

I've been thinking the last few days about improving my code agent harness in codex.

So I just added a visual-qa harness that uses Playwright to diff baseline/pr feature screenshots and already feels like a massive upgrade.

Also seemed to help quite a bit with debugging UI issues by giving a skill combo of screenshots + db debugging with Supabase mcp + Next mcp logs.