# "Just use Claude bro" is good advice. But it only gets you so far. — retained record

> Retained reddit record from reddit:AI_Agents.

Canonical: https://highsignal.app/data/records/c1a77c05f2f9704f

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:AI_Agents)  published 2026-10-23 22:31 ingested 2026-10-24 02:35

[original source ↗](https://www.reddit.com/r/AI_Agents/comments/1xds3zy/just_use_claude_bro_is_good_advice_but_it_only/)

## retained text



```
I'm a hardcore coder with AI, for 6 months I've spent like 12h/day non-stop vibe coding. By API rates my spending is often close to $50k/mo (60B tokens/mo).

Initially I was just using Claude. I quickly realized that skills and tools that people develop and promote are **dead after 2 months** when Claude Code gets updated.

Workflow-orchestration stuff was obviously for people who have **unlimited tokens or unlimited compute** of weak models (such models are too dumb to do good job in single run so they need a smart setup around them; but I'd argue that paying for Claude and just delivering quality code is a better investment).

But there is a limit of where "just use Claude" gets you and it **comes down to three things**.

1. At some point, **you want co-workers to stop asking you questions** about the project. You want them to instead be able to interact with your agent, maybe even suggest changes, bug fixes and contribute to project as well.

At this point you realize, that Claude running on your computer is not exactly what they should interact with. Many people are ok with that and just stop optimizing there. But if you actually want to keep maximizing efficiency, **connecting your agent to others is the way to go**.

2. **Connecting Claude (i.e. Claude Tag) to everything does not scale endlessly**. The more connectors it has, the easier it is for it to confuse the job, project and module and make a (potentially disastrous) mistake and simply making it's work more and more token-costly.

To solve that, you will want to split one agent into project/domain-specific ones, which is why people like **Grok Bot** so much, because that's good decision and even with weaker model like Grok's delivers good outcomes. It also comes with important benefit of **reducing vagueness**, because it by design **forces user to actually choose some agent** and guarantee scoping.

3. **You don't want to be stuck on Claude**. Nor Grok. Nor Codex. Right now is probably the best time ever to be just using Claude sub - they're ahead and giving best value for price. If you're **Grok Bot** user - enjoy your time as well. Both still let us use compute for 1/10 of real price (don't take my word for that, do the research).

Don't be fooled. **Enshittification will happen** \- usage will go lower, models will get dumbed down, prices will rise. It will be mitigated by better models efficiency, routing and so on, but even if prices go up just 2-3x **it's still important to be able to make a jump** to another provider.

**Tooling and setups you run better be model-agnostic** \- even if technically switch might not be complicated, the habits will hit you and your coworkers hard - given company "politics", it will cause **permanent lock-in** driven by fear of change.

\---

**"Just use Claude bro" is a very legit place** to stay at. I decided to make a jump, as there weren't any tools solving issues I mentioned above at that time and I happen to be responsible for improving how we all use AI in company.

Now not only things like **Grok Bot** with **Grok Team Bots** get close to these expectations, but also few dedicated builders (including me) already hopped on these problems and work 24/7 to deliver whatever their users dream of and inform them.

**Just ask open-source devs to work for you** on whatever expectation you or your company has. They'll build it. They're looking for feedback like miners for a gold.

\---

I'd be happy to hear if you agree or not on my 3 points or see more worth adding?
```



document fetched 2026-10-24 10:21 · canonical https://reddit.com/r/AI_Agents/comments/1xds3zy/just_use_claude_bro_is_good_advice_but_it_only
