Short-form insights you can read in under 5 minutes.
Structured planning mode for code generation mirrors what I’ve been doing with my sequenced Plan-Do-Check-Act prompt cycle (https://lnkd.in/effgt9zm ). I’m adjusting my prompts to work within the plan/build modes. Seems like it’ll be smooth as butter.
You can’t tack production requirements onto rapidly produced code. The age of AI hasn’t changed this fact.
Agentic code agents can rapidly prototype applications in hours. This is amazing. These prototypes can quickly help people test the viability and value of a feature or implementation approach.
Production code is a complex mix of such details that require scalability, safety, security, deployability, test coverage, quality evaluation and stakeholder sign off, and optimization for performance, troubleshooting and maintainability. Production code takes orders of magnitude longer to create than prototypes.
The disposable prototypes agents generate usually can’t even be incorporated into such systems without a rewrite. Production software requires a more human centric and expert led approach from the start.
Finally, the speed of delivering software to users is not directly driven by the speed at which the code itself is produced. Code authoring is the middle of a value stream and is not the sole bottleneck in building effective software. AI can also speed up investigation and learning. It can facilitate communication and documentation. In the hands of qualified, experienced teams, AI tools can speed up actual delivery of quality production software by 10-30%.
This gain is worth shooting for but it is an iterative, evolutionary practice improvement. Not something you get out of the box as part of a software license.
Advice? My daughter is moving to the Irish Republic on a visa that allows her to work.
She’s a recent graduate of Worcester Polytech in Robotics Engineering. If anyone out there has ideas on where she might look for engineering related work, technical design and drafting, please DM me!
My learning mindset is driven by humility about my depth of understanding, which extends borderlessly into self-reproach. They are an inseparable essence. One gift.
I believe it's crucial to understand that current AI systems lack true accountability. While they can simulate decision-making, they do not possess moral agency or the ability to comprehend ethical judgments. This creates a dangerous void where responsibility is blurred. We must remember that AI informs our decisions, but we, as humans, remain accountable for its deployment and impact.
I am a failure who has achieved more than I imagined.
I am an ethical human who questions whether I’ve made a positive impact.
I am a happy man who struggles with sadness.
I am a discontent who is at peace with his choices.
I am at any moment any of these. Each wrapped in context without which it disperses.
Each fact eternal. Yet ephemeral and subjective.
Real and nothing.
It’s a privilege to take time away from paid labor. Uncommon in history. Not common enough in our time. Something our parents fought for. Something under threat as the powerful reset the definition of work. The first aspect of a vacation to be grateful for is vacation itself.
Using Anthropic Claude models for coding:
Opus is strong on analysis. Industry benchmarks show it’s strong at coding, but I find it ambitious and rangy. It drifts beyond its mission when struggling, and will charge down supposition paths without grounding.
Sonnet is good at tasking out work and quite good at following through on well-defined tasks, but can be short-sighted in terms of pattern and method reuse.
The current version of Haiku falls short. It shows its age with anti-patterns like aggressively duplicating code and hardcoding values, particularly when spins on a task.
So, for now, I live in Sonnet. I use Opus mostly for analysis on big problems. I’ve stopped trying to use Haiku for now. Would love to hear from someone who is getting decent results in complex code with one of the smaller models.
“In general, there are to(sic) places that LLMs shines - 1) in creating new hypothesis and ideas to test and 2) in creating micro tools to help answer questions rather than trying to provide answers.”
i.e. Using LLMs to surface and present propositions of what legacy code does for humans to evaluate and make decisions over.
The broad conversation is still a funhouse mirror reflection but there are some people describing the strengths and weaknesses of the technology as I actually experience it. So, some hope.
I do use code gen extensively to code but even then it is accelerated learning of both the existing code and unfamiliar dependencies that provides alot of benefit to my developer experience.
Llms are fantastic at summarizing and reformation human language and I get a ton of toil reduction benefit outside of writing code.
I was feeling time pressure and rushing to troubleshoot with my coding agent yesterday and did many of the things that my deliberate practice tries to avoid. I didn’t make it:
So, it partially fixed the problem (yay!), then it obfuscated the work by “fixing” things that weren’t broken (boo) and then, for dessert, it applied breaking “fixes” on top of what it had done (sadness).
I started over.
I’m heading off for two weeks in Japan, my mother’s birth country. For the second week I’ll be hiking with Mark Rickmeier and members of his Walkshop community through an abbreviated Shikoku temple pilgrimage.
I wish all of you opportunities in your life to test yourself, experience new people and places, walk among your ancestral roots, remember lost parents, spend time in your own thoughts, and commune with like minded individuals.
Even if it brings with it a likelihood of travel anxiety, sleepless nights, and joint pain. And those small embarrassments that dive right into the deep pool of childhood fears.