At this point, all of the AI coding assistants are in the same neighborhood. Decent at "advanced autocomplete", OK at code generation sometimes, and most are somewhere in the process of incorporating code context mechanisms. But what's next? Agentic behavior? Something else?
My pet prediction is that we will see the emergence of a new programming language that's designed for use with AI and can be translated to a variety of popular languages. (Or if we're cursed, just javascript )

been leaving comments on a bunch of the AI agent launches this week and noticed the same question keeps coming up in different forms - not "is the agent good enough" but "what's the one thing it's flatly not allowed to decide for itself, no matter how confident it sounds." voice agents that hand off emergencies to a human, coding agents that won't push to main, ones that won't touch payment or lock systems without a human sign-off.

Curious what your actual hard line is, the one thing you'd never let an agent attempt unsupervised even if it swears it knows what it's doing. and just as interesting - has an agent ever actually tried to cross that line on its own, or is it still theoretical for most of you?

Scroll through Product Hunt for two weeks, and you'll see a new voice AI agent, dubbing tool, or "AI receptionist" almost daily. As somebody who's been building in this space and observing it for the last 2.5 years, here're a couple of insights I gathered over time:

Why so many?
Voice is the one interface nobody has to learn. We spoke long before we wrote. Keyboards, apps and dashboards all needed training, but talking doesn t. So once quality got good enough post-2022, adoption didn t need a change in behaviour. It only needed a threshold to be crossed across latency, naturalness or say, interruption handling.
The demand was also already sitting there. A lot of business still runs on phone calls: support, collections, appointments, follow-ups. They re expensive to staff and hard to scale. Also, during last 2 years, building demos got cheaper and cheaper. With decent APIs for speech, LLMs and telephony, one person can wire up a convincing agent in a weekend. This is good for innovation and building use cases for something as niche as HVAC service renewals support.
But what's good for innovation is also the reason for the noise. Buyers are the ones who pay for that noise. Every two-minute demo looks the same. Everyone says 'human-like' and 'enterprise-ready', and almost none of it can be checked before you pay for a pilot. The cutthroat market is trying to undercut each other's prices to zero.

I added a dependency last month because it solved a problem in twenty minutes. Saved me a day, at least.

A week later I needed to change one small behavior in that dependency. Not a big change. A small one. And the library was not built for it. So I forked it, and now I maintain a fork, and the fork is slowly drifting away from the original.

I wrote a README last month. Not because I had to, but because I was putting off the actual work and documentation felt like progress.

It got shared more than anything I have ever built.

Makers: when you change your UI, what goes stale first: your demo video, your help docs, or your onboarding emails?

For us, keeping tutorial videos aligned with the latest UI is an ongoing challenge. Even a small interface change can make an older recording confusing when it shows a button or workflow that users can no longer find.

When I started I thought I knew exactly what people would care about. I had the features planned out and a pretty clear idea of what the product should become. Then actual users started asking for things I never expected and some of the things I thought were important barely mattered to them. It was a little frustrating at first but it also made me question how much of the original plan was based on assumptions. How much do you let user feedback change the direction before it stops feeling like your original idea?

It feels risky to spend time building something just for yourself when everyone says you need to validate a massive market first.

But the truth is market research usually makes you copy what everyone else is already doing. When you fix an annoying bottleneck in your own day you build something with a very sharp edge that nobody else thought of. Sharing those little tools is how our best products actually started.

Have you ever turned a private script into a real public project?

Trying to guess what a generic market wants always leads to the most boring software. Whenever I try to build something based on market research I end up overcomplicating every single feature just to please a hypothetical user who doesn't even exist.

But the tiny scripts I hack together on a Sunday just to fix my own messy workflow always turn out to be the stuff people actually want to use. Building purely out of personal frustration keeps things so simple.

Do you usually release your internal side tools publicly or keep them private?

It's so easy to mistake a launch spike for long-term product market fit.

We tend to pour all our energy into the flashy project because it validates our effort immediately with likes and visits. But the quiet products that solve a real, recurring pain point are the ones that actually survive.

I lost months pushing a hyped tool that churned everyone, while a boring utility tool was slowly building a loyal user base in the background.

Are you planning to shift your focus back to the second project full-time?