AI alchemist, creator, maker • personal views only
🇺🇸
Joined March 2009
- Tweets10.4K
- Following1.6K
- Followers6.6K
- Likes13.2K
This advice from Jeff Bezos is still relevant in the AI era. "I almost never get the question: 'What's not going to change in the next 10 years?' [That question is important] because you can build a business strategy around the things that are stable in time."
So ask yourself…
What constants are most likely to perist for 10+ years?
1. Entropy slows things down. LLMs can tolerate more disorder than humans, but order will always beat chaos.
2. Comprehension is important. If you build for human understanding, an LLM will exhibit total mastery over it.
There's speculation that humans won't need to read/understand code in the future. Not sure I agree. Readability/elegance vs performance is a false dichotomy.
A stack like Ruby + Falcon + Roda + Sequel + Async + Roundhouse + Spinel may just offer the best of both.
Agents re-send the convo each turn, so a file read costs size × remaining turns. I audited 45 days of sessions: 50% of reads pulled whole files, 63% chained multi-turn reads.
Now my agents batch partial reads. Measurably cheaper, faster, and more accurate over longer sessions.
Some traits of software engineers that prove helpful for vibe coding are hubris and skepticism. Keep applying your natural instincts. Never trust what the AI says or does. Push hard for better, more elegant, and more performant solutions. It can always be done better.
AI dynamically creating mini-programs and using them as tools is where we're headed. A naive version of this is the Bash pipelines and one-off scripts widespread today. But it's inefficient. Multi-turn round trips, slow, context bloat, etc.
This approach also satisfies the iconic, "Look at all the things I'm not doing." And expands to, "Look at all the time, tokens, and energy I'm not wasting."
I'm surprised there aren't more discussions about Campfire compiled with Roundhouse + Spinel.
- No code changes required (still Ruby)
- Very competitive in the benchmarks
- Winner in terms of effort/energy/tokens
AI will improve all ecosystems. A rising tide lifts all boats.
I love the term "Vibe Gate" - related to the recent rewrites from A→B
The lazy face value acceptance of most readers. The expose of false comparisons that drove a one sided narrative.
Buckle up! We're in for a future of endless "Vibe Gates" from all quarters.
It's easy to get AI to produce what it already knows, but it takes skill to get new/novel/better outputs that extend beyond pre-trained priors.
I'm finding disingenuous bad faith takes for hype driven marketing purposes increasingly off-putting.
Now that code is basically a commodity, the moat is data capture/creation/curation, preferences, taste, and personalized tooling.
