user34283

The designing and implementing of a code harness in your workflow can be as simple as running something like /skill-builder.

You prompt for what you want it to do, and it will write eg. python scripts as needed for the looping part, and for example use claude -p for the LLM call.

You can build this in 10 minutes.

I don’t use a cloud platform, so I can’t comment on that part. I‘d say just run it on your own hardware, it’s probably cheaper too.

I use the Codex desktop app.

In the GUI I can see the context indicator and usage stats.

It also makes it easier to jump between conversations and see the updates.

Sometimes I use Claude Code or opencode in the terminal, and my experience is much poorer compared to the Codex desktop app.

Velocity without direction?

I‘m personally directing and QA testing every feature.

I don’t know how socioeconomic concerns, product strategy, and mental health are a concern for me here.

I‘m having a great time with my project and it’s been the most fun I‘ve had in many years of building.

It’s in-app on iOS.

Ironically, already another user in this comment section was concerned about the security of my nonexistent backend.

But it’s good to know, I was not previously aware that video processing on the backend is a common source of vulnerabilities.

I would be impressed if you implement it in an hour with the following features:

- webview fallback with canvas capture for codecs not supported in the default player

- detecting blank frames and diff between thumbnails to maximize variety

- UI integration to visualize progress and pending thumbnails, batched updates to the gallery

- versioning scheme and backfill for missing/outdated thumbnail formats

Honestly, a day seems rather optimistic to me. Maybe if I was an expert for this platform and would have implemented a similar feature before, then I could hope to do it in a day.

If I had to handwrite it and estimate it for Scrum at work, I‘d budget a week.

In my experience Codex barely writes any comments, despite my attempts to encourage it in the AGENTS.md.

I feel pretty good about the odds of attackers exploiting security holes in image editing functions my app does not have, in order to enter my also nonexistent datacenter or vpc.

On the other hand, I don’t need to review carefully every line of code in my thumbnail generator and associated UI.

My nonexistent backend isn’t going to be pwned if there is a bug in the thumbnail generation.

After the QA testing on my device, a quick scroll through of the code is enough.

Maybe prompt „are errors during thumbnail generation caught to prevent app crashes?“ if we‘re feeling extra cautious today.

And just like that it saved a day of work.

In my opinion you are just wrong.

It’s an absolute game changer, and it can now multiply your productivity fivefold if it’s a solo greenfield project.

Maybe half a year ago it was as you said. You had to wait for the agent to finish, you had to review carefully, and often the result was not that great. You did not save a lot of time.

Now I can spin up 3+ parallel conversations in Codex, each in a git worktree. My work is mainly QA testing the features, refining the behavior, and sometimes making architectural decisions.

The results are now undeniable. In the past I could not have developed a product of that scope in my free time.

That is what is possible today. I suspect many engineers have not yet tried things that became feasible over the last months. Like parallel agents, resolving merge conflicts, separating out functionality from a large branch into proper PRs.

Other comparisons:

About equal to a major iOS update at 8 GB x 1.5B.

Netflix and YouTube together are perhaps around 200EB/month.

It’s a reasonable expectation from a clearly successful and competent engineer who is using the latest tooling.

Who is to say that it’s wrong?

In a large codebase it‘s probably next to impossible to get people who fully understand the code to explain it to you with unerring accuracy.

AI can get a pretty good picture, near instantly, whenever you need it.

It’s not just competent-sounding, it is reasonably competent, and certainly very useful for tasks like that.

And there’s some truth to it.

I expect tomorrow’s models will be so much more capable that we will happily pay more.

But if not, we will still likely get today’s capabilities or more for cheap.

I don’t see a realistic scenario in which the AI genie is going back into the bottle because of affordability.

It seems like wishful thinking by people who dislike the new paradigm in software engineering.

How do you reconcile these ideas with the fact that cheap open weight models are only slightly behind the state of the art?

If anything, I would bet that next year you could get today’s flagship performance for significantly cheaper via an open-weights model.

I looked into this long ago, and imo it doesn’t look as bad as you say.

Musk downplayed his correspondence and willingness to meet with Epstein to the point where you could argue Musk was lying, yes.

However, he did decline an invitation to the island in 2012/13, at first because Musk was looking for a party and thought this would be a peaceful island experience. Eventually Musk declined because of logistics.

Kimi K2.6, the leading open-weights model, is $0.95 input / $4.00 output / $0.16 cache.

That’s about 1.45x more expensive than K2.5 from January.

It is around 5x cheaper than GPT 5.5.

I don’t think that makes much sense, surely as a billionaire you don’t need to consort with Epstein to meet women around 25 years old.

That link seems to report on the same single WSJ article that mostly alleges workplace power-balance issues, referencing unnamed women, none of whom have come forward to publicly accuse Musk of misconduct. It‘s also fairly thin imo.

Maybe Musk‘s conduct is more gross than I believe, but at this time I‘ll not jump to conclusions.

Luckily inference is cheap, and other providers offer efficient models.

It’s only going to get better in the future.

I never heard about the horse related thing, that’s interesting, thanks.

A long history? Another search tells me that apart from the mentioned accusation, there is only one WSJ article alleging sexual conduct with SpaceX employees.

You asked why I take Musk‘s side in these discussions; it’s because I don’t think he’s a pedophile.

Nothing I‘ve seen seemed convincing to me, and the arguments made online often were so laughably inaccurate and exaggerated as to border on blatant slander.

My searches have not turned up a result showing that Musk "claimed he had never corresponded with Epstein".

Can you source this? If not, can you explain why you did not check it before you posted the inaccurate claim?

On Artifical Analysis it shows only Kimi K2.6 and Mimo V2.5 Pro as better.

Those models are 1T parameters total and 30B or 40B active, this might make abliteration impractical.

About Musk, yes, there is correspondence. The only confirmed meeting appears to be a 30 minute visit at Epstein's house together with Musk's wife at the time.

As for photos you mention, a quick search tells me there is one photo of Musk and Maxwell at a 2014 Vanity Fair Oscar Party.

I find most commentary on here and other platform like Reddit extremely exaggerated compared to what is actually confirmed. Users seem hellbent on linking Musk to pedophilia-related allegations.

It helps if you have a monopoly on app distribution for half of all phones, or video streaming.

Then you can afford zero support and still take 15-30%.

Does it do anything that GitLab does not?

With what I heard about GitHub Actions, the GitLab CI pipelines should be much better.

Not that I haven’t shot myself in the foot with GitLab pipelines on numerous occasions.

Hundreds or 10k+ users?

I imagine requirements and integrations may differ a lot. I have seen many incidents with a large instance.

I am using the Codex desktop app without the pi harness and my experience is quite different.

5.5 has been a noticeable improvement over 5.4, solving more complicated issues and faster too.

5.5 does not use a huge amount of my session limits with the $100 plan.

I use multiple conversations in parallel, all on xhigh effort with Fast on (2.5x consumption), and it’s still enough for me not to switch off Fast.

It also runs my tests, but I did not use TDD apart from sometimes telling it to cover an issue in a test before fixing it.

To evaluate the legal risks of using AI generated code, let’s consider how many lawsuits there have been over these concerns.

Inadvertent copyleft license violations: probably 0 lawsuits

Competitor copied your software, you could not defend your rights in court because it was made with AI: probably also 0

Users of agentic AI for software development: >10 million

The thinking here seems pretty clear to me.

We also pay $300/month for Azure Desktop VMs.

We are paying for tens of thousands of those machines, although everyone knows they are stupidly expensive and incredibly slow.

I will see if that becomes a blocker.

There was one feature/screen that Codex built in a single 5k LOC file.

It was still perfectly capable of developing the feature and it was working as expected.

I had it break it down into multiple files, but if I wouldn’t have seen it during the MR review, I would not have noticed. The large file did not seem to degrade the performance of the agent.

One, thus the git worktrees.

You might think that this would lead to a mess with merge conflicts, but the agent can resolve them automatically.

I added an instruction to AGENTS.md so that before handoff it fetches and rebases, resolving conflicts if needed plus rerunning the tests.

At this point it’s undeniable for my use cases.

After I discovered how to use git worktrees in Codex to work in three conversations in parallel, I am able to build apps with a scope that simply was not realistic before.