I’m finding myself thinking longer about the code I write and the features that get developed, than I did before the agentic coding era.
Slower development, better features# #
With AI now writing close to 100% of my code - and me reviewing and steering 100% of that code - I spend a lot more time on the thinking part of writing software. Writing the initial spec - mostly an incoherent braindump of an idea - and judging an initial iteration. Opening it in a browser, experiencing the UI and UX.
From then on, it’s iteration after iteration. Because generating code is now cheap & fast, I spend way more time on trying out new approaches, new workflows, and nitpicking the details. Where in the past, a feature idea might have taken 1-2 weeks to get deployed, it’s probably taking me 2-3 weeks now from start to deployment. But the end result, to me at least, is always better than what I would have done 2 years ago.
Now that I’m not too occupied with the writing part of code, I can spend all my time on thinking.
Before, the thinking would occur while coding, but I much prefer this new way of software development.
## More in parallel[#](#more-in-parallel)
I’m using [Polyscope](https://getpolyscope.com/)
as my “IDE”, and I constantly have 5+ worktrees active. Ideas or feature requests or bugs that are being worked on at the same time.
Each feature takes a little longer to completely work out, but I’m working on more at the same time, all the time. In the end, I’m shipping more software than ever and I’m confident that what I ship meets our standards of coding and reliability.
## Voice > Text[#](#voice--text)
Ever since I switched to voice dictation using [Wispr Flow](https://wisprflow.ai/)
for my prompts & coding, my *perceived* productivity has jumped up once again.
While this blog post is still written in *the old way*, just by hand, most of my actual work is done speaking these days. For a blog post, I want to be mindful of my wording, sentence structure, coherence, etc.
But an LLM will “just figure it out” based on a messy prompt, and I just iterate from there.
Even more remote work# #
Now that dictation has gotten good enough to be my primary input channel for work, I’m no longer tied to a laptop. For the past few years, I’ve always worked from home, but now it’s even easier to work when not at home.
Out for a walk and you get a new idea? Start a dictation, make lengthy notes. Doing some gardening and you get a brilliant new idea? Braindump it by speaking.
To me, this helps me relax more. I can write down an idea straight away, instead of having to keep it in my head for however long I’m not at the desk. More mental freedom.