logo New Idea

February 2026 Retro

Ego death required.

Hello, loyal reader.

Forget everything I said in my last retro, because I sure did!

I did not in fact focus on the boring bits. Instead, I focussed on the exciting bits. Yup, you guessed it, AI!

Instead of facing the immediate reality of my latest project (i.e., finding customers), I decided to face the larger reality of my career as a software developer. I went against every entrepreneurship lesson I've learned by starting development before capturing customer intent. I treated my latest project as an experiment for building with AI.

By this point, a lot has been written about people's recent experience with AI and the somewhat sudden surge in excitement around using AI agents for software development (since the release of Claude 4.6 and GPT 5.3 range of AI models). That excitement has finally reached my desk and I'd like to document my thoughts on the matter.

I don't think I'll ever hand code again

AI models have reached a tipping point and are simply too good, too fast, and too cheap to not be using them to produce code. So much so that I feel guilty for hand-coding anything, especially during my freelance software dev time, when I know an AI agent can generate the same code in seconds.

AI rate-limits are limiting my productivity

If the future of my software development career has shifted from hand-coding to agent-coding then why the fuck would I settle for being rate-limited during a productive work-day? That's like my IDE limiting the number of characters I can type per day, and I would never settle for that.

As a second-order effect, the participation fee to a career in software development, and by extension tech entrepreneurship, has increased from simply having access to a laptop and an internet connection, to also include a subscription to a capable coding agent. And in the same way a low-end computer or slow internet connection can limit my productivity during the day, a lower access tiered coding agent can limit my progress.

A rising tide lifts all boats

In a post-AI world, if everyone gets a 10x advantage then no one gets a 10x advantage. The most responsible thing for me to do is to avoid being at a 10x disadvantage.

If you can't tell, I'm clearly hyping myself up to upgrade to one of those fancy $200/month AI plans.

Scoping my fears down to an upended career

The future is looking both exciting and scary. On a good day, my fears dampen my excitement. On a bad day, my fears paralyse me.

To avoid ruminating on a future I cannot control, here are a few of my current best-guess beliefs about LLM-based AI:

  • AI is an incredible tool that will greatly expand human capabilities, especially in the digital world.
  • AI won't become conscious (anytime soon, if ever).
  • AI's ability to recursively self-improve will mostly be guided by humans for specific outcomes and be naturally contained by the limitations of our world (e.g., cost, infrastructure, systems, benefits to humanity). Basically, I doubt our world is conducive to sustaining the technological singularity.
  • AI's ability to know which line of code to change to achieve some outcome is improving so much that we can consider ourselves (i.e., human software developers) bested at this task.
  • Humans still seem to have an advantage at figuring out what software to build and why it should exist. Some have referred to this as "taste". This advantage is most beneficial to tech entrepreneurs.
  • The role and responsibilities of a "Software Developer" or "Software Engineer" will likely merge with that of a "Product Manager". However, I think the name of this new role will converge to "Software Engineer" as the focus shifts from writing software to engineering an outcome.
  • AI will enable more software to be built by people of varying skills, outside of software development.
  • People and businesses will continue to limit the scope of their responsibilities to things they care most about, and outsource the rest.

So assuming AI plays by my rules (!!!), I will still have a job and I'll be handsomely paid to do it.

Hidden work

The part not documented here is the mandatory ego death required to allow myself to explore and feel curious about a technology that does the favourite part of my job, better than me.

Fortunately, my entrepreneurship journey had already begun this work on my identity years ago by teaching me that the code is the least important part of building a SaaS business.

Conclusion

The truth is that building software products with AI isn't brain-dead simple (yet). It still requires a certain finesse, perhaps a combination of software engineering, product management, and intuition.

My AI workflow experiment is still ongoing. While the project feels far from production-ready, "far" might only mean a couple of days instead of weeks/months.

Thanks for reading! If you're interested in what happens next, I'll email it to you next month.

Content snacks

I recently found myself needing to spend a total of 12 hours on the road, over the course of 3 days. I do not recommend.

However, I do recommend these podcast episodes which kept my mind engaged: