Comments URL: https://news.ycombinator.com/item?id=47777583

Points: 3

# Comments: 1

- Pay attention to your emotions and body. When you're tired, or scared you are making no progress, stop. Think. How can you delegate more? How can you drive the agent without your input? No micromanaging, the golden rule. If there's discomfort in this process that's more than just momentary, you're doing it wrong. I really believe that. Agents are the ultimate unlock. Think, then free your time to do other things, while the team works. Done well, it's incredibly liberating and satisfying.

- Use Copilot. 39/month for Copilot Pro+. 1500 Premium Requests. Oh look I'm all outta my "5 hour/weekly" limit after being a little too chatty with Opus, or seending Codex down a little too deep of a rabbithole. Fret not - the backup solution, ol Trusty Copilot is the way. Oftimes in recent days ye may have felt the CLI was doomed, buggy, crap. But trust me, Copilot has improved rapidly over the last few weeks. To me, much less buggy. And no limits. Only Premium Requests. And all the models you want, at least that I use. It's really good and useful.

This is not a post about model choices, or tech stacks, or technology in general. It's just about things I've found to get the most out of my tool, while protecting my mood and health. Disclaimer: there is one tech recommendation at the end. No affiliation, just pure utility passing on from one to another.

(Heck, idea: maybe I should just create a blog that let's me do what this simple 3 text box submit does? Anyhow, add that to my list of 25+ ideas rn).

It's not just about getting the work done: it's about feeling good when, and while, you get there. Getting the work done is important, but feeling good, is necessary. This is life, after all, we should be enjoying it. Here's what I've found works in this age of AI dev that remove unnecessary sources of suffering:

- No micromanaging. Agent's think, and change their mind. Interjecting too often not only costs tokens/requests, and can steer them off course, you can be pre-empting their reasoning which would have arrived in the right place. Just because it's not how you would have thought about it, resist the urge to interrupt unless it's an obvious disaster. But then, your spec should have been better. But there's even better reasons to not micromanage, and they are entirely personal, which I'll get to later.

- Verbal specs. Sounds loose, casual, hand-wavy, half-assed? Not at all. I'm at my best when dictating ideas verbally, pacing around. Typing is perhaps for me too much mentally-wired and muscle-memoried into short internet comments, scattered attention, and programming itself. Too much in the weeds, not enough Forest. Verbal for me is all Forest. I can lay it all out, remember it all, and hit all the things I want to if I just can talk it out for 10, 20, 30 minutes. Straight into ChatGPT, who then just cleans it up into a one shot agent spec. Works, every, time. A verbal waterfall.

- Strong invariants and crazy comprehensive e2e tests. With strong invariants baked into that verbal spec, and ambitiously comprehensive e2e (create a clone of production infra, fully populate it, devise a convenience script to push and deploy updates across the fleet, ensure you hit these metrics, pull logs on failure, full headless automation tests for UX, hit these other metrics, collect stats, keep a dev-log and document everything in that, lab-book style, like an experimentalist, etc, etc etc). The scale of te4sting I never would have done myself, agents happily scale those mountains to build gigantic (but not unwieldy) test suites that are not only awesome as health checkups, they are what the agent needs to iterate the fixes you want. They also mean you can close the loop on full autonomy, with strong invariants, a full e2e, the agent can just work until green. You don't need to be anywhere need it. Full delegation. Bliss, for me.

- No micromanaging. Cannot stress this enough. It's a source of major physical and emotional suffering for me. I hate being stuck at the laptop, eeking out progress, lured by the liberating productivity, while feeling stymied by the retarded progress. Ultimate stress out. I just cannot take it. I cannot do it anymore. I could manage myself through lengthy sprints back in the ol' days. But micromanaging agents, is the ultimate "hate this", for me. I feel os much better to just give the clear task, and let it cook all day.

- Let it cook, don't watch the pot boil, unless you feel like mild entertainment or learning a few things. It is educational, occasionally, in small doses, to get a meta picture of the parts of this "craft"/profession: thinking deeply, coding changes, testing, iterating. A watched pot never boils. Somehow this was an aphorism of all ancestor members of my immediate family. Maybe I was an impatient kid, I don't know. But I recall this a lot. In this case, it's true. Resist the urge to interrupt, let the agent cook. Let the pot boil.

Remainder 957 word overage in comment.

Comments URL: https://news.ycombinator.com/item?id=47748747

Points: 1

# Comments: 3

The idea is knowing what to try first today saves a bit of time.

I made a MacOS system 7 web desktop UI with real web browsing: https://win9-5.com/macos/

A re-imagining.

I find the listed side effects don’t happen for me besides occasional flush/blush. Which at my age is more like youthful vigor.

Caffeine is is 1,3,7-trimethylxanthine, pentoxyfylline is 3,7-dimethyl-1-(5-oxohexyl)xanthine.

Good effects are: sustained mental clarity, focus and energy with a smoother more stable baseline than caffeine’s bursty performance; good sleep, but strangely you can also stay up, if you prefer; feeling similar to “after exercise”. Half life is listed as under 1 hour, but beneficial effects can be felt for half a day after 400mg (a standard dose). So maybe there’s something like metabolite dynamics occurring here too.

This ends my erowid/hive style “trip/nootropic” report ;)

Comments URL: https://news.ycombinator.com/item?id=47718289

Points: 1

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=47701478

Points: 1

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=47691573

Points: 2

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=47685116

Points: 1

# Comments: 0

But i guess you know more than i do

I wonder if the US had struck when momentum was high during the popular uprising, it could have being self sustaining, with arms and logistics setup to feed the resistance advance.