Code's last stand

Does o3 spell the end of hand written code?

Last Friday I mentioned o3 (OpenAIs new model which ranks in the top ~0.2% of human coders). Over the past few years, AI has started changing software development. But o3 threatens to turbo charge the rate of change. This model is different - it appears to offer multiple step changes of improvement. And many in the industry believe o3 is just the beginning - and there’s plenty of room for further improvement. The next few years are shaping up to be a wild ride.

Writing code by hand is likely to become largely obsolete within the next two years. Some folks may chose to write by hand for fun; others for learning. And hand written code may hang on for longer for secure/ trusted code where there are regulatory/ political reasons to have a human in the loop. But the vast majority of code will be generated by AI.

• The current AI tools - Github Copilot, Devin, Replit - will be obsoleted and little more than short-lived stepping stones. The age of workflow will be with us. Give a model requirements and it’ll do the rest.

• High-level language development will cease. Currently there are 30-50 new languages created per year of which ~1 gets widely adopted. With AI doing the legwork there is no need for any more languages.

• The models may pivot to writing directly in machine code; Claude is already happy to write (basic) tools in machine code. We use high-level languages because our human brains aren’t capable of maintaining sufficient context from machine-code. But increasingly large context windows are likely to solve that problem.

• Software will become a commodity. Which means the value of existing codebases will plummet. When I can reimplement a modern, right-sized version of your application for negligible cost then why do I buy your product?

We’re about to enter a golden age of innovation and progress. An age of the democratization of software development. An age where everyone has access to top-level quality software engineers. Where anyone can build their own custom tools and programs. Why struggle using Excel to solve a business problem? Just get o3 to build a custom tool for you.

And it’s already happening. Ask Claude to analyse a spreadsheet and it is likely to write custom Javascript to parse the file. Claude is writing custom, throwaway code on your behalf - and without you even asking.

Another example from this morning: I wanted to modify a photo to make it look like one side was torn. A year ago I would search Google & YouTube to try to find a tool/ technique to achieve the effect. Today I get Claude to generate a custom tool and save an hour in the process. For now I’ve got to realize I can ask Claude to solve the problem by writing a tool - by default it wanted to give me instructions on how to achieve the effect with PowerPoint. But how long before Claude offers a tool as the default solution? Six months? Less?

And it’s not just development that improves. Support will also improve - the AI tools are delighted to help and support you. They are always friendly. They never make you feel stupid. Compare that with traditional helpdesks…

Where does this lead?

Building any type of machinery requires technical drawings. The drawings explain how each part is made - and how it fits with all the other parts. Back in the 1980s drawing was done by hand. Engineers spent countless hours hunched over drafting tables. Each design change required redrawing entire sheets. A complex project could take months. Then CAD - Computer Aided Design - arrived.

Despite initial resistance from traditionalists who claimed CAD would reduce quality and destroy jobs, its advantages proved overwhelming. Design cycles dropped from months to days. Changes that once required complete redraws took minutes.

Within a decade, manual drafting largely disappeared from commercial use. New firms, unburdened by legacy practices, outcompeted established players. The industry transformed from drafters who used computers to computer experts who could draft.

We're witnessing a similar revolution in software development. Like CAD, AI will dramatically reduce development time and democratize creation. And like CAD, early adopters will shape the new paradigm while those who resist risk becoming irrelevant.

• Many existing tools are going to rapidly become obsolete. Stackoverflow, Github, VS-code - I wouldn’t be investing in them right now.

• Defining requirements and architecting software will remain important skills - for now.

• Low-level software roles will disappear starting with the lowest skilled. Over the next few years we’ll switch to senior engineers leading teams of AI agents rather than teams of humans.

• Productivity will significantly increase. And time-to-market will similarly drop.

• Expect far more innovation in the market - there will be hundreds of new entrants, many of whom survive only briefly.

• Remember the iPhone “There’s an app for that” tagline? It’ll get transformed into “Make me an app for that”. It’ll be easy to get Gemini or Siri to build tools to solve your specific problems.

What to do?

We’ve likely got 4-5 months before o3 hits the market. And even then it’ll take several months for folk to figure out how to use it. But even if it only delivers half of what’s been demonstrated so far then we’re in for a wild ride. Whether we like it or not, now is the time to prepare for the coming wave.

Think about how best to future proof yourself. What skills will still be useful in 2 years? Architectural, requirements, people skills - they should all still have value. But learning a new development language? It’s probably too late to be of much value. Nor would I invest time on Advent of Code…

• If you’re early in career (or at school/ university) then be aware that many of the skills you are learning won’t be of practical value in a few years. Architecture/ UI/ project management skills will be more future proof than low-level technical skills - focus on those if you can.

• If you run a development team think about how o3 will change the way you approach software development. How can you keep abreast of the changes? How quickly can you pivot when change appears?

• If you run an organization what does this mean for your long-term success? The value of your software assets will erode - and you’ll be lumbered with cumbersome, legacy processes plus staff who are resistant to change. Can you adapt fast enough to survive?

Even if you feel shielded by working for an organization that is stuck with the old ways, be aware they may be usurped by new entrants. AI will significantly lower the cost of entry and leave bloated, inefficient legacy companies at a significant disadvantage. Inertia may give certain industries a few additional years. But ultimately the reduced costs are going to drive change whether we like it or not.

The software development revolution is inevitable. Like the transition from manual drafting to CAD, it will fundamentally reshape how we create. Some will resist, clinging to traditional methods. Others will embrace the change and thrive. But one thing is certain - those who adapt early will help shape the new paradigm.

The question isn't whether to prepare for this transformation, but how quickly you can position yourself to benefit from it. The future of software development is approaching faster than anyone expected. The time to act is now.