Reading Room

You are building AI-leveraged ventures right now, so this is worth sitting with before the next idea. A Reddit essay, sparked by a reaction to the Jubilee debate video 'Andrew Yang vs 20 AI optimists,' lays out a simple, uncomfortable argument: once anyone can type a request into an AI and get working software back, the software business built on answering that request stops being defensible.

In the debate, the optimists argued that AI makes it easier to start a business, with one example being: 'I have this great idea for an app. Now I can finally make it.' The author's response to that example is the whole piece in one line: if all it takes to build the software is typing 'build me a software' into ChatGPT, why would a customer pay someone else to type that same request, instead of typing it in themselves.

The author says this is already happening: more software companies and freelancers are leaving the industry and shutting their businesses down, because anything they build can be instantly recreated by AI. The claim is aimed specifically at builders whose entire product is 'I used AI to make X for you,' not at software businesses that have some other source of value.

The diagnosis is the useful part. The author describes these AI entrepreneurs as people who see 'AI makes it easier for me to do X' and stop there, without taking the next step: if AI made it easier for you, it made it equally easy for everyone else, including your own customer. Once a capability is easy for anybody to reach, it stops being worth paying for, because scarcity, not usefulness, is what lets you charge money for something.

The author is explicit about being pro-AI, not anti-AI, and predicts the same pattern will hit every industry, not just software: so many things become so cheap to produce that prices collapse toward free for nearly everything, except raw materials, and even those get cheaper over time. The piece closes on a blunt line: 'If you think AI is going to make YOU rich, forget it.' The point is not that AI has no value. It is that using AI is not, by itself, a reason anyone should pay you, because your customer has access to the exact same tool.

For your own ventures, the test this argument suggests is simple: if the whole pitch is 'I used AI to build this for you,' ask whether the customer could just use AI to build it for themselves. If the honest answer is yes, the business needs a second source of value, something AI alone does not hand to everyone equally, before it is safe to build around.

This is worth a look because it is a free upgrade sitting inside ChatGPT right now, and it solves the exact problem that kills most AI-made graphics: the picture changes every time you ask for a small edit. OpenAI released Images 2.6 across ChatGPT, Codex (its coding tool), and the API (a way other software can plug directly into the model). The headline change is that images can now move. You can turn a brand mascot, a product shot, or even a rough sketch into a short looping GIF, and the model holds onto the same character, product, and brand details as you make edits, instead of redrawing them slightly different each time.

The scale here is real. People already create 4 billion images a week across ChatGPT, Codex, and the API, more than half a billion a day. Images 2.6 generates those images 50 percent faster than the previous version. In the API, developers get two model options: GPT-Image-2.6 Flare and GPT-Image-2.6 Sunburst. Flare is the fast one. At low quality it can produce an image in as little as six seconds, and it is tuned for layouts like posters and product shots. Sunburst is the detail one: set quality to medium or high and it holds characters and products more consistently across a run of edits, at the cost of speed.

A new setting called complexity controls how much planning the model does before it draws. Low complexity suits quick edits, paired with whatever image quality the task needs. Medium or high complexity is for busier scenes with many elements or constraints, like a crowded poster or a multi-step product shot. Splitting speed, quality, and planning into three separate dials is the detail that matters if you ever script this through the API instead of clicking around in ChatGPT.

The motion feature is where this gets genuinely useful for anyone making quick branded visuals. You can animate a brand mascot for a campaign, turn a product photo into a short attention-grabbing loop, build reaction GIFs or memes for social posts, or sketch out an animated storyboard before committing to real video production. OpenAI's example customer, a startup called Town that builds an AI work assistant, used Images 2.6 through the API to give its characters more personality in day-to-day interactions. None of this needs a finished design to start: a simple sketch can become an animated character.

Two other changes matter if you ever edit images programmatically rather than by hand. Image markup, already in ChatGPT, is now in the API too, so you can point at the exact spot on an image you want changed instead of describing it in words. And exports now include SVG (a vector image format that scales cleanly for web, print, and apps) alongside the usual flat image files. Images 2.6 is live today in ChatGPT and Codex on every subscription tier, on desktop, mobile, and web. GPT-Image-2.6 Sunburst and Flare are both available in the API now, with pricing posted on OpenAI's site, and OpenAI published a safety write-up covering how it tested the update before release.

Outside your usual reading: a developer built a small CSS trick where pixel art characters walk across a webpage and react to its layout. Resize the browser window and the little walkers treat each line of text as its own floor, dropping from line to line and falling off the end of short lines, while boxes, buttons, and images block their path but the text itself never does. It has no business point at all. It hit 244 points and 48 comments on Hacker News, a reminder that building things can just be fun.