HTML All The Things - Web Development, AI, and Developer Careers · Matt Lawrence and Mike Karan

Trying Codex For The First Time Was… Confusing

·47 min·1 clip

AI tools are getting more powerful, but the UX might actually be getting worse

The episode opens from a blunt place: AI can already build a lot for you, but the first encounter with a new coding assistant still matters. The hosts describe an early run with Codex and quickly feel the tension between capability and usability. Before long, the tool is running commands, asking permission to execute things, and burning through usage credits. That sequence becomes the starting point for a broader complaint about the current state of AI developer tools. The hosts are not saying the tech is weak. They are saying the experience around it can be messy, and sometimes messier than people want. A big part of the discussion is the friction at the point where the user is supposed to get started. They talk about how Codex was not available on Windows in an app form the way they expected, and how the setup path already felt a little off. That leads into a conversation about folder selection and whether a small looking interface choice is actually a serious usability problem. The hosts do not treat those details as trivia. They treat them as the sort of thing that decides whether people keep going or back out. The episode also spends time on the economics of these tools. The hosts talk about the assumption that a $20 monthly subscription should cover the experience people want, and why that mental model can break once usage based systems enter the picture. They separate Codex usage from ChatGPT usage and note that hitting Codex limits can stop access for a while. That matters because the cost pressure changes how a consumer or prosumer thinks about experimentation. It is one thing to try a tool. It is another to feel every prompt burning through a budget. From there, the conversation becomes less about one product and more about the shape of the market. The hosts talk about people not wanting to subscribe to several AI coding products at once. They point out that consumers may simply stay with the tool they already know, even if better alternatives exist. That makes adoption feel sticky in a very human way. Convenience wins over novelty more often than product teams like to admit. There is also a recurring theme of construction zone software. The hosts use that feeling to describe where AI tooling is right now: powerful, active, and still rough around the edges. They keep coming back to who is paying, how much mental overhead the tool adds, and what the normal user is supposed to understand before they start typing. That practical framing keeps the episode from drifting into abstract AI commentary. It stays anchored in actual usage. The hosts also circle back to the idea that this kind of tool may land more naturally with people already comfortable in developer workflows than with casual users. If you already live in developer tooling, some of the friction may feel manageable. If you do not, the experience can feel like a lot very quickly. That tension gives the episode its shape. The tech is exciting, but the human experience still lags behind it. Another thread in the conversation is social perception. The hosts reference the chatter around high daily costs for some prosumer AI setups and how that kind of number changes the emotional reaction before the tool has even had a chance to prove itself. That reaction is part of the story. People are not only evaluating what a tool does. They are evaluating whether they trust the pricing, the limits, and the workflow enough to keep using it. The overall pace stays conversational and exploratory. The hosts think through the problem in public rather than presenting a polished verdict. That gives the episode a lived in feel, especially when they are trying to separate what is technically possible from what is actually pleasant to use. The result is a discussion that is less about Codex as a brand and more about the current stage of AI tooling as a category. By the end, the episode has made its main point clearly: more power does not automatically mean a better product. In fact, the hosts suggest the opposite can happen if the interfaces, limits, and onboarding do not keep up. That is the tension running through the whole conversation. The excitement is real, but so is the annoyance. The episode also captures a very current developer mood. People are willing to try these tools. They are just not willing to pretend the rough edges do not matter. That makes the conversation useful even when it gets a little messy, because the mess is part of the subject. It sounds like working developers trying to make sense of a tool while it is still changing under their feet. A final thread is the practical uncertainty around access and expectations. The hosts keep returning to what users think they are buying versus what they actually get when usage caps and separate subscriptions enter the picture. That mismatch sits at the center of the episode. It is not just about one tool feeling confusing. It is about a whole category still learning how to be understandable. The conversation closes in the same spirit it starts: useful, skeptical, and focused on the lived experience of using AI to build software. It is less interested in slogans than in the awkward moments where the product meets reality.

As heard by us

An early look at Codex that ties power, friction, and separate usage limits into one clear UX problem.

Codex is the starting point, but the larger subject is the distance between raw capability and daily usefulness. The episode uses an early look at OpenAI's Codex to make that case plainly: it can run commands, ask for permission, and burn through usage credits fast, yet still…

Read the full review in PlayNext →

Why you'd press play

You want the Codex experience without the hype gloss.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from HTML All The Things - Web Development, AI, and Developer Careers