Steve (Big Red) Bishop 👨‍💻

Bishop's Law revised Parallel Systems: R = O(α log C + β) Concurrent Systems: R = O(α C + β) The addition of O() uses Big O notation for asymptotic upper bounds, while "log" reflects logarithmic scaling in optimized parallel implementations, versus linear in naive ones. These changes indicate that design improvements can reduce (but not always eliminate) resync complexity; naive async systems approach linear or worse scaling, limiting efficiency at large scales.

@unclebobmartin As I tell fellow developers all the time, you don't have to change what exists, you can always create new to replace old.

Mutation testing has a dark side. Not only does it consume rather large amounts of CPU and wall time; but it makes is much more difficult to remove old behavior and replace it with new, "better", behavior. Those extra tests do their job of stabilizing the behavior very well -- perhaps a bit too well. Caveat Emptor.

@allenholub We aren't programmers, we are problem solvers.

People tell me that using an LLM to create code will rot my programming-language skills. So? We work in a changing landscape, and agility is essential to survive. My goal is not to be the best Python (or whatever) programmer on the planet. My goal is to produce the most valuable, high-quality product in the shortest time possible. To do that, I want to use the most effective tools available. That might be a programming language, an LLM, a combination of the two, or something else entirely. Toolsets are constantly evolving. Keeping up is part and parcel of being a developer. I need to be skilled with the tools I'm using right now, not the ones I used in the past. People who rest on their laurels quickly become unemployable. I guess the difference between the critics and me is that I've never seen myself as just a programmer (though I'm pretty good at that, if I do say so myself 😄). I'm a developer. I develop products and tools. To do that, I need to know how to program, but I also need to know architecture, product discovery and refinement, systems thinking, testing, TDD, UI/UX, and a host of other skills, including communication and process improvement. Every one of those things is integral to what I do, and most of them are not impacted by the LLM at all. As for those rotting coding skills, I still need to code to use the LLM effectively. I need to read and understand the code, refactor it when needed, and write things by hand that the LLM can't or won't write. With the tool, however, I can work faster. When I moved from assembly language to C, from C to C++, from C++ to Java, from Java to Kotlin, from Kotlin to Python, the earlier skills indeed rotted away. I didn't much care. I can get them back easily enough if I need to. I'm with Sherlock Holmes, here. When Watson told him that the Earth goes around the Sun, he said that now that he knew it, he'd do his best to forget it. "What the deuce is it to me? … You say that we go round the Sun. If we went round the Moon, it would not make a pennyworth of difference to me or to my work."

Be honest, is this relatable??

Accurate. Agentic coding isn't just the future, it's the present. It's a wild bucking stallion that if tamed, will lead to expedited results. I'm implementing ideas that before I just shelved because it would take up too much of my time. Innovation is going to explode for the next decade.

@alexbunardzic This is why I have a coding guidelines document I require each LLM to review before implementing code changes. It has specifics such as Outside-In TDD, clean code, vertical slice architecture, trunk base dev, etc. I routinely update it with any little gotcha's.

Right now the biggest problem with AI-assisted software development (for me) is spending too much time comparing the development flow as it unfolds when using different models/platforms.

@unclebobmartin I would be very curious to know the dependency metrics of a clean architecture build vs vertical slice architecture build of the same app.

I had codex build a dependency checking tool that finds cycles, enforces dependency boundaries, and calculates dependency metrics. It found cycles. Who knew? So now we are breaking cycles using dependency inversion. I am so happy.

@unclebobmartin Told ya.

Codex is definitely faster and probably smarter than Claude. I'm pleased with it so far.

More of this, please.😂

@unclebobmartin Don't forget, it is often saying one thing but doing another.

I asked codex to apply a clean architecture solution (components and dependency inverted boundaries). It wrote up a nice plan describing the components and the boundaries. Then I asked it this: › would the result of executing this plan make it easier for you to work with this system? Why? • Yes. Because it would reduce ambiguity and hidden side effects, which are the main things that slow safe changes: 1. Single mutation boundary With apply-command!, I can trace state changes in one place instead of chasing swap! across many files. 2. Clear dependency direction If domain/app no longer depend on atoms/UI/filesystem, changes are localized and lower-risk. 3. More pure functions Pure transitions are easier to reason about, test, and refactor without breaking behavior. 4. Better failure isolation Ports/adapters make bugs easier to classify: domain rule bug vs UI adapter bug vs persistence bug. 5. Faster, safer edits When invariants run after commands, regressions are caught immediately, so iteration speed improves. So it helps me work faster, but more importantly it makes changes more predictable for everyone on the codebase.

I reject your assertion. The entire field of software engineering evolved under the fundamental constraint of the human ability to manage complexity.

@unclebobmartin "Claude, remake Facebook. Make no mistakes."

I cheated. I looked at some sample code and, although it was pretty good, I thought it could be better. So I told Claude to improve it.

@abhijitwt Fortran

guess the language he codes in?

Have you ever noticed that the worst people keep getting promoted? You’re sitting in a meeting thinking, How is this person in charge? That’s not bad luck. That’s a system working exactly as intended. Here’s the truth: A lot of workplaces don’t promote the most capable people; they promote the least threatening ones. The people who move up fastest aren’t always driving results. They are the ones who keep leadership comfortable. - They don’t challenge decisions. - They don’t surface problems too clearly. - They make things feel calm, even when nothing is actually improving. Real competence creates friction. When you are genuinely good at your job, you expose gaps, weak processes, bad decisions, and poor leadership. And that makes people above you uncomfortable. So instead of developing strong leaders, many organizations quietly sideline them. What gets rewarded instead is loyalty, predictability, and the ability to manage optics. Once that pattern starts, every layer protects the one above it. Promotions stop being about skill and start being about safety. And over time, the people doing the real work either burn out from carrying everyone else or they leave. - That’s how mediocrity becomes a culture. - That’s why leadership can feel hollow. - And that’s why, when you look around and wonder why so many managers seem unqualified, the answer isn’t random. It’s structural. They didn’t fail upward by accident. They were rewarded for not rocking the boat.

@unclebobmartin Maybe use a different model than Claude?

The more I press on Gerkin style specifications for AI, the less I trust them. The AI will happily claim to be passing all the Gerkin tests, while in reality implementing them to silently pass while testing nothing. Nothing can be trusted other than direct human inspection at the code level. And that inspection must be deeply suspicious of the lies the AI will tell and the ways that it will hide it's misbehavior.

This is the fuzzy world that engineers of all disciplines grapple with. Making exacting order from inexacting chaos. AI is a layer of abstraction that interfaces with the chaos and pretends to provide order. Which is why it behaves precisely like a beginning engineer.

AI CEO SAID SOMETHING HONEST looks under covers oh dax, yeah checks out

@mysticwillz Ask Copilot if it knows of any recent job postings that fit my experience.

Your API keys just leaked on GitHub. 50,000 requests in the last 3 minutes. AWS bill climbing by $200/second. The attacker is still active. What is your immediate lockdown sequence?

It sure is fun paying $17/mo to run one prompt then wait another 5 hours before you can run your next prompt to fix what the previous prompt broke. #claudethings