RobIW

15 yrs SQL + BI. Building the tools I want that don't exist(or they cost and arm and a leg). AIST protocol

[RobIW @RobIW_dev]

@neogoose_btw The cost savings. As soon as I stumbled upon one of your tweets I knew it was good

[RobIW @RobIW_dev]

switched to gpt models and codex 5 weeks or so, iirc, and the rate limits being so much more generous given most work can be done on low or medium has unlocked parallelisation I never had with claude models.

not to mention the resets of course :D

[RobIW @RobIW_dev]

the parallelisation going on now that you can use resets at your wish is awesome

[RobIW @RobIW_dev]

the more I try gpt 5.5 on medium and low the more it dawns on me it can do so much and it's only about the prompt and workflow almost like higher reasoning is mostly just to make up for gaps and flaws(obviously it's for much more than that but in terms of general code dev)

[RobIW @RobIW_dev]

@ZhihuFrontier @Zai_org How does it perform on deepswe?

[RobIW @RobIW_dev]

Not claiming my workflow is on the level of fable but the general principle of using varied agents for planning and analysis and designing the solution/implementation is indeed incredibly good at producing results that one model simply can’t do.

Even if you only use two agents of the same model you’ll get results that are literally more than 100% better

[OpenRouter @OpenRouter]

Introducing the Fusion API, the smartest compound model in the market.

Fusion achieves Fable-level intelligence at half the price.

How it works 👇

[Dennison @DennisonBertram]

Major life hack:

DeepSeek in the Claude Code harness can also build and drive workflows, at a fraction of the cost and Opus 4.6/7 quality.

I've got it running over 250 subagents in a workflow in adversarial reviews.

Pennies on the dollar.

Use my tool "Deep-Claude"

[RobIW @RobIW_dev]

@mark_k @ArtificialAnlys was obvious from the initial 24hrs of using it

but I was going off of feeling, glad it's now not just a feeling

[RobIW @RobIW_dev]

one of the biggest improvement to your workflow you can do is not import others skills and repos etc(although that is also great obviously) but to run a retro on your workflow whenever an issue pops up

yes after any issue

after a while and a few of these retros you will iron out all the gaps in your workflow to the point where your work is mostly one-shot except for issues that are very complex and once you get to those types of issues your workflow will be iron proof

We basically had to restep a 7-step process to repro this issue but most of my flows in apps are rarely this long(by design)

[RobIW @RobIW_dev]

@morganlinton yeah fable 5 was the first model where I just straight up asked it "what model do you think is suitable for this task?" and then asked how to rework my main router skill to have this q come up early when creating tasks

[Tibo @thsottiaux]

Heard your (amusing) feedback that it was at times annoying to receive a reset of your Codex usage without warning.

Next time we press the button you will get to choose when it actually applies. Happy codexing.

[RobIW @RobIW_dev]

@JohnPippett @bridgemindai yeah the list goes on :D

[RobIW @RobIW_dev]

@xeophon They want you to burn as many tokens as possible doing repetitive things

That’s their business model

[RobIW @RobIW_dev]

@sabari_246 @claudeai Yeah you need to set the effort level according to how difficult your task is. Most tasks are handled perfectly fine on medium or low or even ask it if a cheaper model is sufficient

[RobIW @RobIW_dev]

This is what fable should have been. Fuck yes

[Pankaj Kumar @pankajkumar_dev]

GPT-5.6 : Coming Next Week

- OpenAI Chief Scientist Jakub Pachocki said GPT-5.6 will be a "meaningful improvement" over GPT-5.5 (via The Information).
- A new anonymous model called "Kindle" has appeared on Design Arena, likely linked to the previously leaked kindle-alpha GPT-5.6 checkpoint
- GPT-5.6 is competitive with Claude Fable 5 while potentially being available at a much lower price point
- GPT-5.6 is expected to bring major improvements in reasoning, coding, agentic workflows, vision, and frontend generation
- OpenAI is also considering significant token price cuts as competition with Anthropic continues to intensify

[RobIW @RobIW_dev]

“That’s strictly more capable than picking two. The machinery cost is leases + epochs (to detect divergence) plus a reconciliation pass (to resolve it) — both of which the old handoff already specified. So your design always had the trifecta; I read §15 too literally.”

Another fable moment. The spec already says what I want but somehow, when asked for general opinion on the spec, it tries to impose restrictions that limits functionality…..

[Matt Pocock @mattpocockuk]

Steps to become a senior programmer:

1. Install my /teach skill

npx skills add mattpocock/skills --skill teach

2. Create a new working directory on your laptop

mkdir junior-to-senior
cd junior-to-senior

3. Kick off your coding agent in the directory

claude

4. Copy this prompt

/teach me how to be a great strategic programmer. My opinion is that AI is eating 'tactical, on-the-ground' programming. The day-to-day work of a developer involves not only coding, but also planning, QA, codebase design, and much more. I'm interested in learning the strategic skills - that, in a previous era, would take me from junior to senior - but in this era are table stakes.

5. Paste it into the coding agent

Below is an example of what the first output will look like. I used Opus 4.8, medium effort.

6. Continue working with the agent until you're a senior

[RobIW @RobIW_dev]

just had a 15 minute back and forth with fable 5 about why it's perfectly fine to build this and within terms of service but still get flagged and moved to stupid 4.8

meanwhile I had gpt 5.5 medium build it and now it's done.