Software Development in the Time of Strange New Angels

DAVEGRIFFITH

This must read article explores the impact of AI on the software industry, from the simple perspective that developers cost roughly $150/hour, and as a result code is scarce and expensive. The processes we have built up within our industry (Devops, Agile, Testing Pyramids) are optimised around protecting that expensive and precious time. An interesting viewpoint.

But what happens if the cost of producing code collapses from $150/hour to near zero - and created in seconds? We’re on the cusp of that becoming a reality, or at least a reality within certain contexts.

Dave shares his own experiences, which mirror my own, that the autocomplete tools that were our first taste of AI, have been eclipsed by something much more powerful:

“And that was the moment that Claude stopped being a tool, and started being a colleague.”

These are the angels referred to in the title:

“If you know just what code needs to be created to solve an issue you want, the angels will grant you that code at the cost of a prompt or two.”

So is this the ‘end of days’ for our industry? No, not yet. Most organisations lack the testing, architecture, deployment, and product judgement to exploit this new reality.

The parting thought from this post really resonates with me. The real future constraint may become wisdom: deciding what should be built when almost anything can be built quickly and cheaply.

Introducing GPT-5.1 for developers

OPENAI.COM

And this week’s model release … is another from OpenAI.

GPT-5.1 builds on GPT-5 which was released three months ago, where one of the most notable new features was its ability to route questions to either a fast model, or a reasoning model. Something I’ve found to work very well in ChatGPT as a conversational AI.

This release is an incremental improvement, allowing you to manually turn of reasoning for latency sensitive tasks, cache prompts for improved reasoning and follow-up questions. It also adds a couple more tools for patching and shell automation. And finally, the SWE-Bench (verified) scores have improved by a few percentage points.

ByteDance’s Volcano Engine debuts coding agent at $1.3 promo price

TECHINASIA.COM

It’s pretty safe to assume that most of our AI consumption is heavily subsidised, with much of the bill for our “by the token” frontier model usage usage footed by the investors. At some point I expect there will be less investment flowing and we will start having to foot the real costs - or these companies will realise they are providing a service that competes with a $150/hour human cost, and want to retain more of the value they are creating!

But that’s tomorrow’s problem. If you’re unhappy paying your $50 per month, ByteDance are offering a $1.30 subscription!