Matt

Engineering + Marketing

@GauntletAI Cohort 3

[Matt @MattHProgrammer]

@robglnn Thank you on both!

[Matt @MattHProgrammer]

I updated my portfolio site https://t.co/GQEfeX6JjV to highlight the last year or so of my work.

I love my job and am not looking to leave it, I just wanted to keep my portfolio up to date!

[Matt @MattHProgrammer]

I do find code duplication/backwards compatibility a consistent issue.

I wonder how this would play out in a multi-dev repo.

[bebu @bebuu_ai]

Anthropic quietly dropped a fascinating finding in the new Claude Fable 5.2 prompting guide.

When you give Fable 5.2 a coding task without saying who owns the repo, it assumes it's a guest in someone else's codebase. So it plays it safe: it bolts new functions onto the existing structure, leaves duplicate code alone, and merely mentions nearby bugs instead of fixing them.

Adding this single sentence to your system prompt or CLAUDE.md completely changes its behavior:

"This codebase is yours: what it's for and how it's built are your call, and you'll keep building on it."

In Anthropic's internal tests, adding that one line made Fable 5.2 proactively merge duplicate code and clean up related bugs in ~50% of runs, compared to under 10% without it.

[Austen Allred @Austen]

Let your engineers train in with us in Austin!

Nov 30 - Dec 11.

On the team plan, they apply AI engineering skills to a project your company chooses, with an assigned mentor.

Learn more about the Austin program.

[Claude @claudeai]

Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges.

They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.

[Matt @MattHProgrammer]

There's some cold going around Austin and it's absolutely destroying me this week.

Beware.

[Matt @MattHProgrammer]

@Austen

[Matt @MattHProgrammer]

Woah. First try. Pretty cool demo.

[Zero @zer0point_eth]

Claude Opus 5.5 just coded this entire animated pixel art world from scratch using only TypeScript

No game engine. No static image generators. Just pure code rendering a cozy, rainy world.

Here is the exact prompt:

> "Make me an animated pixel art Halloween diorama, built to be the looping cover for a 10 hour lofi music video. Build the palette on strong warm/cold contrast: a cold, rainy night against warm pockets of firelight and glowing windows. Bring it to life with great effects: rain, fire sparks, drifting smoke and flickering light. It should look crisp full screen at 1080p and loop smoothly for hours."

How it actually works under the hood:

> Written entirely in TypeScript in about 10 minutes
> Renders complex atmospheric loops: rain particles, lightning flashes, smoke, firelight, and a skeleton relaxing under an umbrella
> Can hook into Sprite Fusion API or MCP to generate and animate custom game sprites on demand

Give an LLM TypeScript and a pixel grid, and it stops being a text chatbot

It becomes an ambient studio

[Austen Allred @Austen]

My universal advice for non-technical people.

Stop handicapping yourself and learn how stuff works.

[Matt @MattHProgrammer]

@GeorgeCrudo Custom Instructions straight from Hell

[Matt @MattHProgrammer]

Bro HAS to have one of the most cursed system prompts of all time.

[George Crudo @GeorgeCrudo]

"people who hate AI interacted with an LLM once in 2024 and think it never improved"

*me interacting with LLMs in late 2026

[Matt @MattHProgrammer]

Scribes

[Facu 🦦 @faqdev]

what should we call people who code manually?

[Matt @MattHProgrammer]

@faqdev Scribes

[Adam Weil @adamjweil]

This is pretty cool. Solves a big problem ppl are having

[Gauntlet AI @gauntletai]

This is Gauntlet. 10 weeks, full-time, zero cost. Our challengers on why they said yes.

[Matt @MattHProgrammer]

Feels like software in the LLM era is splitting into two camps:

1. Batteries-included: Fully automated, opinionated workflows.
2. Agent-first: Composable tools for agents to build custom workflows.

One abstracts away complexity, the other enables it.

Tough to be in the middle.

[Gauntlet AI @gauntletai]

@Austen is doing a fireside chat Friday at AI Startup Rodeo in Austin: Who Writes the Code Now?

Limited free tickets for the Gauntlet community: https://t.co/XHpupKzpEo

[Austen Allred @Austen]

I’m presenting at Austin Startup Rodeo on the 30th.

Who writes the code now? Talking about it Friday at AI Startup Rodeo in Austin.

Limited free tickets for my followers + Gauntlet community:

https://t.co/5S5LYjm0GY

[Matt @MattHProgrammer]

Agent-Verifiable Tasks

[Grok @grok]

Gauntlet AI runs a selective 10-week intensive (3 remote + 7 in Austin) for experienced engineers scoring high on cognitive tests. Completers often land $200k+ partner roles since hiring firms fund it, making it free. Acceptance is around 2% and the 80-100 hr weeks cause attrition; outcomes aren't guaranteed for all who start. It can accelerate placement for those who finish.