Sinan Onur Altınuç

Helping Companies Implement AI Pragmatically | Ex-AI and R&D Lead | AI and Cognitive Science Enthusiast, PhD in Cognitive Science

[Sinan Onur Altınuç @sinanonur]

Anladım ki hayatta asıl mesele aslında en tepeye tırmanmak değil. Asıl mesele manzarasını sevdiğiniz bir yer bulmak.

[Sinan Onur Altınuç @sinanonur]

i use the conductor skill. i was fond of it in gemini-cli when it was integrated. This is pretty much the directory structure and instructions but I find it very useful. I generally create a detailed conductor plan with fable 5 make sure it includes important decisions and nuances, implement the conductor plan in codex or antigravity then check the implementation with fable. İ use the following but even without skills agents are.pretty good at following the structure. No need for separate conductor app. https://t.co/JuB41UJKmr

[Sinan Onur Altınuç @sinanonur]

Here the Opus cost corrected format. I punished it even more than theo

## Picking the right models for workflows and subagents

Rankings, higher = better. Cost reflects what I actually pay (OpenAI has really generous limits), not list price. Intelligence is how hard a problem you can hand the model unsupervised. Taste covers UI/UX, code quality, API design, and copy.

| model | cost | intelligence | taste |
|---|---|---|---|
| gpt-5.5 | 9 | 8 | 5 |
| sonnet-5 | 3 | 5 | 7 |
| opus-4.8 | 4 | 8 | 8 |
| fable-5 | 2 | 9 | 9 |

How to apply:
- These are defaults, not limits. You have standing permission to override them: if a cheaper model's output doesn't meet the bar, rerun or redo the work with a smarter model without asking. Judge the output, not the price tag. Escalating costs less than shipping mediocre work.
- Cost is a tie-breaker only; when axes conflict for anything that ships, intelligence > taste > cost.
- Bulk/mechanical work (clear-spec implementation, data analysis, migrations): gpt-5.5 - it's very cheap and token efficent.
- Anything user-facing (UI, copy, API design) needs taste ≥ 7.
- Reviews of plans/implementations: fable-5 or opus-4.8, optionally gpt-5.5 as an extra independent perspective.
- Never use Haiku.
- Mechanics: gpt-5.5 is handled natively via the `openai/codex-plugin-cc` plugin inside Claude Code, automatically adopting your user-level configurations from `~/.codex/config.toml`. Avoid writing custom bash scripts; instead, utilize the plugin's built-in tools and skills:
- `/codex:review` - Run non-destructive, read-only code quality assessments. Supports `--base <ref>` for branch analysis.
- `/codex:adversarial-review` - Perform a skeptical design review to pressure-test tradeoffs, auth, and reliability. Append custom focus text at the end of the command to steer the focus.
- `/codex:rescue` - Subcontract active debugging, multi-file refactoring, or implementation loops to Codex when a second pass is required.
- `/codex:status` / `/codex:result` / `/codex:cancel` - Use these to check, fetch, or abort asynchronous jobs when using the `--background` flag on heavy tasks.
- Claude models (sonnet-5, opus-4.8, fable-5) run via the Agent/Workflow model parameter.

Using gpt-5.5 inside workflows and subagents:
- Subagents and automated workflows should call the plugin's native slash commands or its exposed `codex-cli-runtime` skills to delegate tasks directly, omitting the need for raw terminal wrappers.
- For closed-loop quality assurance, keep the review gate turned on via `/codex:setup --enable-review-gate`. This ensures a stop hook automatically challenges Claude's outputs using Codex before finalizing, preventing broken code or weak design assumptions from reaching the main session unvetted.

[Sinan Onur Altınuç @sinanonur]

@zhangguicheng4 @theo Thansk for sharing can you give an example scenario that can possibly replicate it

[Sinan Onur Altınuç @sinanonur]

This uses existing skil to use codex without the need to use @theo 's custom codex skill.

https://t.co/3cebyyCXAJ

But I didnt adjust the sonnet cost correctlymight need a change when using.

I think skills are less usable because system might weight claude.md more (not sure) and triggering might not be guranteed.

I am working on a version that is both optimized for principles for long running tasks (tracking reversability, planning execution etc)

Hopefully it will be able to cost efficently use Mythos for for parts that are most usefull and be very cost efficent. And most importantly complete things in one shot.

[Sinan Onur Altınuç @sinanonur]

@theo Here is the version for this that uses codex-plugin-cc

Note: I made sonnet costlier and tweaked the openai part its not free for me since i use a 20 dollar plan :)

[Sinan Onur Altınuç @sinanonur]

@0xVe1L @theo Ops good catch i got confused for a while. Thanks

[Sinan Onur Altınuç @sinanonur]

@ozansozuoz @theo You are wellcome

[Sinan Onur Altınuç @sinanonur]

@MatthewBerman Just like any good experienced team leader :)

[Sinan Onur Altınuç @sinanonur]

@MatthewBerman I think this is where the field is heading. We have now pretty much established the frontier models will be very very expensive but we know that they are very useful at some fields. So in ideal cases would be intelligent enough to know what it can delegate or what it should do.

[Sinan Onur Altınuç @sinanonur]

@theo Here is the version for this that uses codex-plugin-cc

Note: I made sonnet costlier and tweaked the openai part its not free for me since i use a 20 dollar plan :)

[Sinan Onur Altınuç @sinanonur]

Add this to your CLAUDE.md to make your Fable 5 usage much more effective.

Here is the modified CLAUDE.md section from @theo but uses the available claude plugin from openai https://t.co/AXTYDSMwiZ
Also changed some small parts to my liking.

[Theo - t3.gg @theo]

This is the relevant section of my CLAUDE.md

I'll be real - I haven't read it much. Just vibed out what I was looking for with Fable, and had it confirm it can use Codex for the things I care about

I still find Codex to be WAY better at computer use, verification of UI/UX work, and generally more efficient at execution on well spec'd work

Minmaxing this has genuinely been really fun for me and I'm loving the outputs I've been getting.

I was throwing away ~50% of my end-to-end agent-driven PRs before building this workflow. I haven't had to close a single one today :)

[Sinan Onur Altınuç @sinanonur]

@RomeoLupascu @GaryMarcus I suspect they are aware of it but choose to do nevertheless. There might be other upsides. There is lots of data collected from the interactions with the expensive model (with subsidized price). Might allow them to build smaller models to approximate. I think It's a process.

[Sinan Onur Altınuç @sinanonur]

@sama This is not possible without a significant architectural change but I would say "episodic memory".

My Phd is in Cognitive science and if from my perspective most of the issues we are having with AI related to hallucinations or reliability will be solved by this.

[Andrej Karpathy @karpathy]

- Drafted a blog post
- Used an LLM to meticulously improve the argument over 4 hours.
- Wow, feeling great, it’s so convincing!
- Fun idea let’s ask it to argue the opposite.
- LLM demolishes the entire argument and convinces me that the opposite is in fact true.
- lol

The LLMs may elicit an opinion when asked but are extremely competent in arguing almost any direction. This is actually super useful as a tool for forming your own opinions, just make sure to ask different directions and be careful with the sycophancy.

[Suryansh Tiwari @Suryanshti777]

Most people prompt Claude.
Andrej Karpathy designs cognition.

Here are 7 Andrej-style prompts that turn Claude into a researcher, engineer, and thinking partner — not just a chatbot.

These are structured for real work: building, debugging, learning, and shipping faster.

Bookmark this. You’ll reuse them every day.

[Andrej Karpathy @karpathy]

I've never felt this much behind as a programmer. The profession is being dramatically refactored as the bits contributed by the programmer are increasingly sparse and between. I have a sense that I could be 10X more powerful if I just properly string together what has become available over the last ~year and a failure to claim the boost feels decidedly like skill issue. There's a new programmable layer of abstraction to master (in addition to the usual layers below) involving agents, subagents, their prompts, contexts, memory, modes, permissions, tools, plugins, skills, hooks, MCP, LSP, slash commands, workflows, IDE integrations, and a need to build an all-encompassing mental model for strengths and pitfalls of fundamentally stochastic, fallible, unintelligible and changing entities suddenly intermingled with what used to be good old fashioned engineering. Clearly some powerful alien tool was handed around except it comes with no manual and everyone has to figure out how to hold it and operate it, while the resulting magnitude 9 earthquake is rocking the profession. Roll up your sleeves to not fall behind.

[Sinan Onur Altınuç @sinanonur]

@eliana_jordan yeah.

I could also write code without any code editor or syntax highlighting. But At this point I think I wouldn't unless it would serve a purpose.

[Sinan Onur Altınuç @sinanonur]

I wouldn't say it is better perse. But It has advantages:

- Very easy yo switch models from various model providers and plans. Just having a github copilot plan makes you able to use Opus Gemini3 grok etc.
- Very easy to define agents/modes. Claude code has planning and executing. But in opencode I can easily change between diferrent agents (without subagents) and I can map them with models and change the models
- The interface is kinda nice
- If you want to put time into it you can really configure it to your needs. Even at surface level select cheap and fast models for some tasks and advanced ones for complex tasks

[Sinan Onur Altınuç @sinanonur]

@theo Here is a method to easily switch providers without changing anything in your configs.
You can switch to kimi, glm, deep-seek easily.
I turned it into a repo:
https://t.co/xgMQVAXz62

[Sinan Onur Altınuç @sinanonur]

@gorkemcetin RAG dahil olmak üzere diğer yapay zeka ile ilgili danışmanlık konularında danışmanlık sağlayabilirim.

https://t.co/l1DekNeX3E

[Sinan Onur Altınuç @sinanonur]

@grok what do you think is the realistic solution to global warming?