gabriel-uribe

Haven't been excited for a Gemini release since December. Wild to see.

Appreciate it! I plan to. To be fair, I've made small updates here and there to this one. Just no overhaul/expansion yet.

Best of luck on the beta & release!

Continuing to work on https://attachedapp.com.

ChatGPT validates your spirals. We help you prevent them.

We're in the middle of v2 over the next few months based on everything we've learned since launching ~a year ago.

As a habitual side project guy for the longest time, it's so satisfying to finally stick to one thing and go really deep. This thread has been highly motivating.

Very cool! I made something similar a while ago for learning Mandarin Chinese. Unfortunately more basic than your Spanish version: https://dailychinesestories.com.

I miss working on language learning tools. My attempts were all in the ~2024 era of LLMs.

We're still in the early adoption phase. Betting against the internet wasn't a great idea, and betting against AI doesn't look like one either ;)

Fwiw, don't buy into all the hype that you're falling behind. Yes, AI does cool things now, but I would say the impact is still unproven past indie hackers or early-stage startups. And a lot of the esoteric setups people have created with things like OpenClaw have become outdated as quickly as they were conceived.

The popular thing is now to setup loops (eg I setup hourly integrations for Claude/Codex to 1) scrape my Linear, claim achievable tasks, and push PRs or 2) do root cause analysis on customer issues that evaded automated filters, to name a few)

Though for me, my setup still feels mundane. I have AGENTS.md, CLAUDE.md etc and a few skill files. These are purposefully light - tons of examples online you can pull from online. Mine are fairly personal to my setup and products.

Importantly, I also allow Claude and Codex to bypass permissions. Yes, there is a risk they wipe my machine. The productivity upside has been worth it, for me (haven't been burned yet, ~9+ months into running models this way, I have backups, use cloud etc).

As far as maintaining quality, one of the most helpful guardrails over the past year, for me, has been requiring my agents to pipe their changes to local reviewers through OpenCode, Cursor, etc agents to have a council of models with different biases reviewing the changes, and autonomously working towards a completed objective. No matter how good Claude or Codex gets, for example, I will probably always want a different model checking its work. Like GLM, (now with 4.5) Grok, Composer.

Several OpenAI, Anthropic, xAI employees, and popular AI engineers post on X and share helpful tips & updates. Highly recommend for keeping a pulse on startups and AI. I haven't found something close, honestly, other than when I spend time in SF talking to people.

Wasn't writing about major companies. That's obviously next, if we follow the trend lines.

I was also under the impression modern AI agents have moved on from just OCR'ing screenshots to leveraging native vision model capabilities.

I remember being blown away by o1-o3 family of models finally stringing together coherent agentic tool calls to write and execute scripts semi-reliably for workloads in the several minutes before they would start hallucinating/flailing. GPT 5 was a bit ahead of that, but barely

Now we take for granted that the latest models can juggle between multiple browser tabs, applications, databases, simulators, docker etc to write, execute, e2e test and deploy full-stack applications over hours managing up to dozens of subagents, relatively untouched, without taking down prod even 1% of the time

Not only this, but in the GPT 5.0 era, agents had 0 taste. Nothing looked good. It was the agentic version of the twitter bootstrap era, but worse somehow. Now, I would argue the average agent frontend beats the average human frontend. This isn't even getting into 3D applications in the GPT 5 era

Anyway, the models now reliably execute more than a human can fit into their own context. It's magic

This season of Silicon Valley is getting spicy

The toggle feels useless in the desktop app. It just changes a few shortcuts?!

Admittedly, I was already using Codex a bit like Claude Cowork. I'm just surprised they decided to merge threads.

Definitely. They had insanely low rates on TTS up until a month or two ago ($4.20/1M) for example, which they only recently started increasing.

As their models get more competitive I'm sure prices will catch up.

Womp. Didn't see this anywhere else.

No longer feels as inexpensive. Will likely just include this in the rolodex of <200k context tasks, like being one of my review agents.

Quite draining, I hope this is not what the future holds.

Not sure about Korea, but Japan's case is more interesting than simply adjusting for population: the government strongly incentivizes new buildings every few decades. There is also a culture of viewing housing as a consumable, which is the part that sticks out to me the most.

I think so too. I'm an early 30s that would love to pay a fraction of the current market rates for a Tokyo-sized unit in SF.

Yep. Counterintuitively, housing in Japan depreciates unlike most of the world

So many chart crimes, but aside from that.

Seems like we're seeing margin get eaten up by companies upstream of the buildout, and the costs have not been fully passed downstream yet. Eg most consumers still get free/cheap AI.

It'll take a few years to see what happens at scale when prices go up and purchasing behavior changes. Hardware, services etc all downstream of the buildout and supply constraints.

This makes sense. Early on most of our support requests were for routine things eg login issues, but now it’s more complex: “this new genAI feature did this weird thing”

Honestly, would love to do that. Main issue is margin. Earning just enough to even consider hiring, but not nearly enough to hire someone great that will stick around. Have tried hiring a few times now, but it hasn't worked out for various reasons.

Kicking the can down the road for now.

Deterministic scripts are awesome, and they certainly power my internal dashboards.

But I'm a human - I will miss things. I maintain too many apps to have entire codebases in memory at this point. Or to continue monitoring all these streams. Logging is cheap - I log as much as possible because an AI will scan it for me.

I just want scoped pull requests to review proactively against the slew of things that can happen in prod that I didn't account for in my specs (again, from logs, customer issues, etc). I discard most of them. That is fine.

I don't miss the days of scraping through logs or dashboards myself to troubleshoot some latency or malformed data issue that I missed conditionals for.

AI is incredible at finding patterns in otherwise benign stdouts, let alone as it cross-references data streams.

In theory, I don't need most of these automations. But for $200/mo? I will happily reduce my cognitive burden on stuff that doesn't impact the core business and make it easier to keep things gliding smoothly.

When the subsidized plans disappear, I will keep these automations going with the best small models that fit on my laptop.

YMMV but automations eat through the $100-$200 plans, which burn thousands in tokens alone.

I have hourly automations for root cause analysis on customer support issues, daily automations for eg log analysis, weekly & monthly automations for KPI tracking & actioning.

I will say, when I was building side projects that were 1) fairly well defined in scope and 2) without users/need for automations it was much easier to stay under $20/mo plan limits. Now I regularly hit weekly limits and need multiple Max plans

Salary has had weak correlation to productivity gains since last century.

This feels like a symptom of the definition of "real work" changing right in front of us. Some people still use AI like a copilot, cleaning up code here and there, maybe writing functions. And at the right scale, this is genuinely still real work.

Others, especially startups or indie hackers, use AI like it were their end-all be-all assistant. "Hey Jeeves, go add Apple Sign In, Google Sign In to our signup pages. Also, investigate why we're not utilizing cached inputs on our AI APIs correctly. And add Maestro flows for every screen in our app. Btw check out posthog, supabase, and Stripe - is our new agent changing engagement or trial->paid conversion rates?"

And 3 hours later, you have all these done. But only if you use the right multi trillion param models.

3-4 cups a day. Was deeply stressed building products with <$100/mo in rev. Couldn't sleep at night from all the caffeine and cortisol spikes.

The ensuing several months were some of the worst in my life (many confounding variables) missing coffee, feeling anhedonic, and inadequately caffeinated/lucid to tackle my problems.

I'm almost exactly 1 year coffee-free (not caffeine free, but significantly less because tea is much less addictive for me).

Also positive in the long-term for me. Fewer digestive issues, less spiky dopamine sensitive or impulsiveness and performance during the day, better memory. I wish it weren't so.

But damn was the 3-6 months of anhedonia awful. I still feel pangs of it.

The toolbox for anxiously attached people, like myself: https://www.attachedapp.com

Most iOS/Android mental wellness apps are trying to be everything for everyone, ie general AI journaling or meditations.

By niching down, we can build the best experience end-to-end for anyone that resonates with these particular emotional challenges.

100% this.

I also have Claude Cowork automations running constantly. As-is, I can't shut down my laptop, and it gets frustrating when my laptop is in my backpack all day because of commutes or travel.

TIL. That's cool. I mostly use Gemini 3 Flash for some background jobs because of the price/perf, but rooting for their models to improve. Competition is good.