eugeneonai

hace 2 mesesdiscuss

[flagged]

hace 2 mesesdiscuss

Great, will it be possible to see it in your profile?

hace 2 mesesdiscuss

The 79% / 67% reduction generalizes broader than IaC. Any CLI agents shell out to (curl, jq, grep, kubectl, gh, psql) burns the same token tax — verbose JSON, free-form text output, agent-composed pipelines. A predicate-flag + compact-output redesign would land on all of those.

  Direct answer to your question: agents-writing-IaC-in-prod is rare today but not zero. I see more "agent reviews the IaC PR a human wrote", which Cost.dev sounds well-suited to since verification runs locally and the agent only consumes the result. Even if the prod-IaC path takes another year, the design pattern earns its keep on every agent-shellout you already do. One question: does the CLI surface its cache state to the agent, or does each invocation start fresh Repeated price-fetches across a single agent run would be the obvious next-tier savings.

eugeneonai

hace 2 mesesdiscuss

[flagged]

eugeneonai

hace 2 mesesdiscuss

[flagged]

eugeneonai

hace 2 mesesdiscuss

Agree on RSS as the right shape — and worth adding the cost angle nobody's quantified here yet. Having an LLM read a 50KB HTML page is ~$0.03 of gpt-4o input. Polling 1000 sources hourly = ~$720/day, almost all of it tokenizing layout chrome the model throws away. RSS-shaped feeds drop that 90%+ because they strip to deltas. The harder blocker is the supply side though — publishers earn pennies per human pageview from ads and ~$0 from agent polls, so unless feeds become licensed paid endpoints, the publisher incentive runs against your "publish an RSS feed for your content" recommendation. Just like that :)

eugeneonai

hace 2 mesesdiscuss

Looks nice. The README pitches model-per-task picking but doesn't say much about context management. In coding-agent loops the full system prompt + tool specs re-send on every step — a 30-step task pays the input cost 30x. Prompt-cache headers catch the static prefix, but the per-step diff (file diffs, observation tokens) isn't cached, and that's often most of the input. Auto-summarizing older trajectory into a state vector saved 40-60% input tokens in workloads I've looked at — could be a useful daemon-side concern since users won't reach into each agent's internals.

eugeneonai

hace 2 mesesdiscuss

[flagged]

eugeneonai

hace 2 mesesdiscuss

I agree, but my situation is a little different.

I try to squeeze my hobbies into the workweek, and it ends up happening after work.

I work from 10 a.m. to 10 p.m., then pursue my hobbies for a few more hours, leaving no time for a "personal life," let alone sleep.

But I can't physically move my hobbies to the weekend. If I did, I'd feel trapped and tied to a schedule. I can't do that. I need freedom, even if it's the same "freedom" that, in my mind, directly interferes with my sleep and, to some extent, my health, which I sacrifice for the sake of my hobby.

But it's important to understand what a "hobby" is. A hobby is something I enjoy that positively impacts my life in one way or another. If your interest (hobby) "satisfies" you, then it will take precedence over sleep, food, and so on. Just like me.

My hobby is creating products to make life easier. Initially, it wasn't monetized, but with the addition of monetization, it now feels like the meaning of life.

I love weekends, I love weekdays, and I love my life with its many flaws and nuances. Thank you for the post. It was interesting, as I've been thinking about the same thing and still do, but I can't do anything about my "hobby" that I can't transfer, even though it could have been a better decision.

eugeneonai

hace 2 mesesdiscuss

Now I'll know where to get new ideas. I'm off to school, guys.

The most original things can truly be learned from people whose thinking isn't yet "proper" to adults. I've always thought about that.

eugeneonai

Submissions

[untitled]

comments