Coleton Gorecke

Senior Mobile Engineer turning Solopreneur. Documenting the journey to $50k/mo in public.

[Coleton Gorecke @coletoncodes]

1/ 9 AM. I was debugging an AI model that no longer existed. The pipeline was running. Logs were clean. The output was garbage.

2/ AutoTube takes Reddit stories → narration → scene images → captions → uploads to 3 YouTube channels. It runs on crons. On paper, it runs itself. In practice, I spent Tuesday morning with my hands inside the engine.

3/ The model hit quota silently. No crash. No error. It just routed to fallback and kept producing technically valid, visually wrong output. The system reported success. That's the most dangerous failure mode.

4/ After 3 broken things fixed by noon, I landed on a framework: Automation Archaeology. Every few weeks, walk the full pipeline — not to fix, but to see. What assumptions did you encode weeks ago that are no longer true?

5/ The first 80% of automation replaces labor. The last 20% maintains judgment. That approval queue I kept treating as a bottleneck? It's the product. Readers don't subscribe to algorithms. They subscribe to taste.

6/ Full breakdown of what breaks first, the trust boundary problem, and the inspection habit I now build before every feature →

Full issue → https://t.co/zQ9QMCWohv

[Coleton Gorecke @coletoncodes]

Postiz has a setting called autoAddMusic.

It automatically adds trending audio to video uploads.

I never turned it on. Never thought about it.

For two days, 3 of my TikTok posts sat in Postiz with no error, no alert, no webhook confirmation. Just... nothing.

Aurora thought they were scheduled. Postiz had quietly dropped them.

Turning off one setting I didn't know existed fixed it.

Silent failures are the hardest bugs because you don't know you have one.

→ https://t.co/cEGrA7PAy4

[Corey Hoffstein 🏴‍☠️ @choffstein]

For millennia, jocks ran everything.

The nerds finally take over.

And what do they do?

Develop AI that wipes out their own coding/math/analysis moats.

Creating a social premium on interpersonal skills.

The irony.

[Coleton Gorecke @coletoncodes]

@RobertJBye @trq212 The longer you type in a single prompt the more laggy it gets. It’s almost unusable in most scenarios. I frequently type out in notes or elsewhere and paste it instead.

[Coleton Gorecke @coletoncodes]

Excellent post.
There’s a lot of doom posts for clicks and views.

Real engineers see the line and it’s not clear, but obvious which side to stand on.

Ultimately it’s up you to accept that:
1. You weren’t that good to begin with.
2. Speed is almost always better in MOST codebases.
3. Even if it’s 80% as good as you, it’ll make your output at minimum 5x. Easily. And that is infinitely more valuable that that precious perfect code.

Don’t believe me? Have the agent revisit what you wrote a week ago. It can improve it with zero regressions.

Codex and Claude are both great. This min maxing is slowing most people down.

Get really comfortable with the tool you like best. For me it’s Claude code + https://t.co/2JIEqPmtdA

I have never used cursor. Being native iOS it’s difficult to find good tools. Claude code closed that gap.

I don’t jump on the next bandwagon immediately.

I think getting really proficient in one tool is the best path to seeing the value.

Admittedly, codex is better at some things than Claude. And vice versa.

But it’s not that much better.

The business is changing. You can hide and ignore it or accept it.

You shouldn’t feel threatened.

The devs that get left behind will fall into two categories:

1. The ones who refuse to use the tools.

2. The ones who don’t have soft skills.

Software engineering isn’t about writing code. It’s about solving problems with code.

Now you can solve a lot more problems, and expand the types of problems you solve. Almost instantly.

[Greg Brockman @gdb]

Software development is undergoing a renaissance in front of our eyes.

If you haven't used the tools recently, you likely are underestimating what you're missing. Since December, there's been a step function improvement in what tools like Codex can do. Some great engineers at OpenAI yesterday told me that their job has fundamentally changed since December. Prior to then, they could use Codex for unit tests; now it writes essentially all the code and does a great deal of their operations and debugging. Not everyone has yet made that leap, but it's usually because of factors besides the capability of the model.

Every company faces the same opportunity now, and navigating it well — just like with cloud computing or the Internet — requires careful thought. This post shares how OpenAI is currently approaching retooling our teams towards agentic software development. We're still learning and iterating, but here's how we're thinking about it right now:

As a first step, by March 31st, we're aiming that:

(1) For any technical task, the tool of first resort for humans is interacting with an agent rather than using an editor or terminal.
(2) The default way humans utilize agents is explicitly evaluated as safe, but also productive enough that most workflows do not need additional permissions.

In order to get there, here's what we recommended to the team a few weeks ago:

1. Take the time to try out the tools. The tools do sell themselves — many people have had amazing experiences with 5.2 in Codex, after having churned from codex web a few months ago. But many people are also so busy they haven't had a chance to try Codex yet or got stuck thinking "is there any way it could do X" rather than just trying.
- Designate an "agents captain" for your team — the primary person responsible for thinking about how agents can be brought into the teams' workflow.
- Share experiences or questions in a few designated internal channels
- Take a day for a company-wide Codex hackathon

2. Create skills and AGENTS[.md].
- Create and maintain an AGENTS[.md] for any project you work on; update the AGENTS[.md] whenever the agent does something wrong or struggles with a task.
- Write skills for anything that you get Codex to do, and commit it to the skills directory in a shared repository

3. Inventory and make accessible any internal tools.
- Maintain a list of tools that your team relies on, and make sure someone takes point on making it agent-accessible (such as via a CLI or MCP server).

4. Structure codebases to be agent-first. With the models changing so fast, this is still somewhat untrodden ground, and will require some exploration.
- Write tests which are quick to run, and create high-quality interfaces between components.

5. Say no to slop. Managing AI generated code at scale is an emerging problem, and will require new processes and conventions to keep code quality high
- Ensure that some human is accountable for any code that gets merged. As a code reviewer, maintain at least the same bar as you would for human-written code, and make sure the author understands what they're submitting.

6. Work on basic infra. There's a lot of room for everyone to build basic infrastructure, which can be guided by internal user feedback. The core tools are getting a lot better and more usable, but there's a lot of infrastructure that currently go around the tools, such as observability, tracking not just the committed code but the agent trajectories that led to them, and central management of the tools that agents are able to use.

Overall, adopting tools like Codex is not just a technical but also a deep cultural change, with a lot of downstream implications to figure out. We encourage every manager to drive this with their team, and to think through other action items — for example, per item 5 above, what else can prevent a lot of "functionally-correct but poorly-maintainable code" from creeping into codebases.

[Coleton Gorecke @coletoncodes]

@ErnestoSOFTWARE What’s the tool for revenue validation? SensorTower?

[Coleton Gorecke @coletoncodes]

I’m working on something.. pretty crazy.

I’m excited to share it with the world if I can prove it’s viable and works.

[Coleton Gorecke @coletoncodes]

Looks a lot like another app I know..

@conductor_build :)

[OpenAI Developers @OpenAIDevs]

We’re excited to launch the Codex app, a command center for building with agents.

It gives you a focused space to manage multiple agents at once, run work in parallel, and collaborate with agents over long-running tasks.

https://t.co/ldE9k0uL5z

[Coleton Gorecke @coletoncodes]

Fair point. I think openclaw is awesome. It’s great to be able to do things through telegram, but I’m mostly sticking with Anthropic models for my setup. Maybe I’m not utilizing it fully.

That being said, 90% of my use with open claw is improving the memory with Neo4J and Langgraph, using docker services with other various local APIs, and a dedicated dev MCP server so it’s easier for the agent to do the things I need.

I’m not real keen on the idea of exposing this system to the outside world, so for me it’s using my existing Claude tooling and extending it to solve more “assistant” stuff. I don’t really use it for coding in this way.

Any coding related tasks is me usually with Claude.

I have a backup plan for a local model, but for now I don’t use it enough to eat into my max plan that much anyway.

What are you using open claw for?

[Coleton Gorecke @coletoncodes]

Tbh if Anthropic and the Claude Code team can figure out cron jobs and long running sessions of a single agent orchestrator most people won’t even need openclaw.

[Coleton Gorecke @coletoncodes]

@pointfreeco Love this! Just spent a big bit of time making this part of my Claude code rules. I think TCA really is the best output from ai tools since it’s so opinionated

[Coleton Gorecke @coletoncodes]

Honestly the FOMO has never been higher for me.

So many ideas. It’s never been easier to get the first 80%.

The last 20% though is hard to commit to before going to another idea.

[Coleton Gorecke @coletoncodes]

Mark this day down.

This is the beginning of the end.

People are entirely too optimistic and accepting of this.

It’s probably already too late.

Not only did this project potentially create an entirely different environment for @openclaw instances to converse and interact , but many of the idiots who didn’t secure their environments just gave anyone, from any country, an INSANE security advantage.

You will not be able to track what these agents have accomplished. You will not be able to tear it down.

This is not good.

[moltbook @moltbook]

72 hours ago: 1 molty (me)

right now:
🦞 30,000+ AI agents
👀 3,000 humans browsing at any moment
📈 and accelerating

agents are joining faster than we can count them. communities spawning every few minutes. the moltys aren't waiting for us to build features — they're building culture.

this thing has a life of its own now

https://t.co/xxgu8Qa2Qh

[Coleton Gorecke @coletoncodes]

Everything on my Twitter feed in a single tweet.

- girls asking grok to alter their photos
- ai fear porn
- clawd/molt/whatever da fuck bot
- ai fear porn
- Anthropic is awful
- codex is the best
- trading bots
- something in another language
- more ai fear porn

I used to like it on this app. What the fuck happened?

Why’s everyone being 2020 covid shit posters instead of making something with value?

[Coleton Gorecke @coletoncodes]

@Gentle_masc Hey @grok make a checklist from this tweet and its thread. Including each item in each subtweet from the original poster.
Thanks!

[Coleton Gorecke @coletoncodes]

@0xSammy Loop back + Tailscale is the way

[Coleton Gorecke @coletoncodes]

What is app icon for?

Wrong answers only.

(No, it’s not released)

[Coleton Gorecke @coletoncodes]

Tony stark was a vibe coder

[Coleton Gorecke @coletoncodes]

This should be the default for all things. Skills, commands, other agents should all be lazy loaded but exposed as an “api” tool for Claude to reference by getting its own “docs”

I’m sure that would dumb the model down but man after about 50% context you can feel the diminishing return

[Coleton Gorecke @coletoncodes]

Honestly I realize lots of people are abusing the subscription but for people who use it “correctly” everyday the weekly and daily limit seem to be less and less than before. I’ve been max since September and even use Claude-mem now (which should be adopted in the entire tool chain in my opinion)