Wisen Tanasa

Many of the software engineers I talked to told me that they practice TDD out of guilt. That's not right, we need a different way to think about TDD. I've been practising TDD for almost a decade, here's how I currently think about it:

@ragunathjawahar I presume the graph tool is crawling files rather than exported modules?

@ceilfors Another interesting fact that I found out today. If we decompose a React application by components, with one component per file, the resulting dependency graph resembles the component tree 😉

• Build • Split • Stabilize

@andrewcairns On agents, I think there's something about the promise of less work that conflicts with the recent findings (agents intensify work).

@ceilfors Oh I'm not sure. Tea and cookies are a good motivator for me, though. Really not sure if this will work for AI agents :)

How does motivation differ when helping agents versus helping humans, including oneself?

"All the tests" implies all the scenarios are knowable upfront, though. My workflow is a bit like this now: 1. Prompt to write tests with all known scenarios (Canon #1 + #2). Commit. 2. Prompt to make all tests pass (First half of Canon #3 + Canon #5). Commit. 3. In a loop: Think/prompt to discover other scenarios, accept/delete, implement. (Canon #2 + Last half of #3). Commit. 4. In a loop: Prompt to refactor. Commit.

@johncrickett @bygregorr Interesting question--all the tests or one test at a time? Nobody knows. Yet.

TDD changed how I write software. Agentic engineering is changing it again. I don't think the classic red-green-refactor loop of TDD makes sense when an AI agent is writing the code. I've been a proponent of TDD since I read @KentBeck's XP Explained. I roughly followed what he describes as "Canon TDD": 1. Write a list of the test scenarios you want to cover. 2. Turn exactly one item on the list into an actual, concrete, runnable test. 3. Change the code to make the test (and all previous tests) pass, adding items to the list as you discover them. 4. Optionally refactor to improve the implementation design. 5. Until the list is empty, go back to #2. I say roughly, as I was never good at sticking to #2. I'd turn two or three test scenarios into runnable tests at a time, often using table-driven or parameterised tests. This process adds value when a human is writing code. But when an AI agent is doing the writing, I think Test First Development is currently a better approach. We can define all the test cases, let the AI write all the tests, then let the AI write the code to pass them. Doing each step one at a time would be a much more expensive way of using the agent. There would be far more tool calls, far more roundtrips to the inference engine, and it would consume a lot more tokens. Then there's the difference between the agent and an engineer. The agent isn't necessarily going to refactor like a human engineer would. I don't think we lose much from this change. We can still guide it after each feature to evolve the design. We can still iterate quickly and get fast feedback on the design and functionality. Are you still doing red-green-refactor when the agent is doing the typing?

Decompose or die trying.

Comfort compounds weakness. Stress compounds strength.

@samlambert We removed material-icons and iconsax, and the memory consumption drops from 15GB to 1.5GB 🤷🏻

am i doing this right?

@eikonne Whoa.

I can't believe my last tweet was back on June 2024! Thinking to get more active again to process my thoughts here. Who is still around?

Better problem understanding leads to better design. This technique from @ericevans0 helps achieve exactly that. read.ceilfors.com/p/concrete-exa…

@andrewcairns It doesn't hurt to grow taller to 2990cm. Imagine the possibilities of what you could achieve!

@ceilfors Height goal, lol You've met me in real life - am I too short? Should I be spending more time trying to get taller? 🤣

My new new tab in Chrome shows me a snapshot of key metrics I care about. Can anyone name all 3...?

@JasonSwett The demo 🤯

@ceilfors If you really want to be impressed... youtube.com/watch?v=T2uEUB…

CALIBRATION SEQUENCE INITIATED

@JasonSwett Sounds like an exciting project, I've got a Lego spike and have not built anything as sophisticated!

@ceilfors It's the (crappy) beginnings of a robot that can play chess.

@andrewcairns @ddd_eu Thank you! Your new video? 🤣

@ceilfors @ddd_eu Loved this @ceilfors - and very timely for me while I'm about to: "cuddle the baby [which is not mine], before I kill it!" 🤣

I attended Eric Evans' Software Design Masterclass held in @ddd_eu, and wrote one of my many learnings here: read.ceilfors.com/p/unspoken-sec…

I'm looking for someone to join us at Aardling to bring better modelling and software design to our clients. If you're a software architect and you have experience coaching teams, building a continuous delivery mindset, get in touch :-) aardling.eu/en/software-ar…