Ian Adera

I am an artist and code is my medium

[Ian Adera @ianodad_dev]

Claude is smart, but its answers can be a wall of text.
One prompt in my CLAUDE.md fixed it. Every reply now arrives in 5 parts:
→ what happened
→ what it means
→ what's risky
→ recommendation
→ next step
I read once and act.
https://t.co/6YGjZqV2h7 #ClaudeCode #AI #DevTools

[Ian Adera @ianodad_dev]

@Hesamation and it consumes 30% more tokens than Sonnet 4.6

[Ian Adera @ianodad_dev]

2 hours into my first AI-agent session: an auth bug I could not explain. I wrote none of it. I accepted the diff because it looked right.

That is vibe coding.

How to use Claude Code as a real dev without losing control:

https://t.co/11IoqKEpSH

#ClaudeCode #AI #DevTools:

[Ian Adera @ianodad_dev]

@ClaudeDevs

[Ian Adera @ianodad_dev]

@ClaudeDevs Yeah, something wrong with this 5-hour limit session

[Ian Adera @ianodad_dev]

@eclarkhalid @ClaudeDevs same here

[Ian Adera @ianodad_dev]

@eclarkhalid @ClaudeDevs experiencing the same issue as well

[Ian Adera @ianodad_dev]

@ClaudeDevs now it seems like like my 5 hour session are draining very fast as well.

[Ian Adera @ianodad_dev]

Fresh week ..

[Ian Adera @ianodad_dev]

@bridgemindai Its back with more limit per session.

[Ian Adera @ianodad_dev]

@antonpme It looks like they have increased limits again. i have been running agents and bearly burning up tokens

[Ian Adera @ianodad_dev]

@strwbcom Enjoy more limits. I think they were switch to the spacex data center

[Ian Adera @ianodad_dev]

Anthropic limits have increased. It's going to be a productive week.

[Ian Adera @ianodad_dev]

@BowesChay

[Ian Adera @ianodad_dev]

I can see the model, context used, session cost estimate, and rate-limit reset. with a tiny status line

The context % is the real win. It tells me when to handoff, compact, restart, or move key details into a spec file.

Less guessing. More control.

Full post → below

[Ian Adera @ianodad_dev]

That has made the workflow feel less like “AI coding for me” and more like working with a small technical team.
The real productivity gain is not just moving faster.
It is having clearer lanes: who executes, who reviews, and where human judgment must stay in control.

[Ian Adera @ianodad_dev]

I’ve been using subagents in my dev workflow, and the biggest lesson so far is simple:

AI becomes more useful when you stop treating it like one assistant and start routing work properly.

In my spec file, tasks are routed based on complexity, which the model helps classify.

[Ian Adera @ianodad_dev]

Simple mechanical work goes to lighter agents. Scoped coding tasks go to executor agents. Deep review goes to stronger reviewers. Risky or architectural decisions stay with me.

[Ian Adera @ianodad_dev]

@NdemoKelvin Claude is so much better. My token run out with claude. But couldn't still use codex.

[Ian Adera @ianodad_dev]

@zackvoell True