Matt Florence

AI Engineer 🤖 ⌨️ 🚀 Outdoor Adventure Junkie ⛰️ ⛷️ ⛺

[Matt Florence @mattflo]

Agentic systems have two fundamental particles: text the model reads, and code it can ask to run. One stochastic, one deterministic.

Skills and progressive disclosure exist for one reason: keeping the stochastic half on the rails.

https://t.co/g1u5UJBUKq

[Matt Florence @mattflo]

Among knowledge workers, tacit knowledge is often a pricing problem, not an epistemic one. LLMs change the economics. Latent knowledge is going liquid.

[Matt Florence @mattflo]

LLM assistance in software creation blurs the line between planning and execution. Sure, execution is streamlined. But more fundamentally, well-crafted plans are becoming executable.

[Matt Florence @mattflo]

LLM coding moves the bottleneck. Look for weakness in three new places: good ideas, human judgment, and clear communication among people building software.

[Matt Florence @mattflo]

Feedback loops grounded in human value judgment have always been critical. In the era of LLM coding, they matter more than ever.

[Matt Florence @mattflo]

I've completed "Playground" - Day 8 - Advent of Code 2025 https://t.co/0OfHgz1AeP #AdventOfCode Whee!

[Matt Florence @mattflo]

I've completed "Laboratories" - Day 7 - Advent of Code 2025 https://t.co/uXLPjEumRs #AdventOfCode Definitely the hardest so far.

[Matt Florence @mattflo]

I've completed "Trash Compactor" - Day 6 - Advent of Code 2025 https://t.co/SJVSwNLsHi #AdventOfCode

[Matt Florence @mattflo]

I've completed "Cafeteria" - Day 5 - Advent of Code 2025 https://t.co/k9pAS7Yq4c #AdventOfCode

Thankfully the optimization I implemented to get partA to work, also worked beautifully for partB.

[Matt Florence @mattflo]

I've completed "Printing Department" - Day 4 - Advent of Code 2025 #AdventOfCode https://t.co/mwAGDEzryb

After day 3 part 2, was happy to find both parts of day 4 a bit easier.

[Matt Florence @mattflo]

I've completed "Lobby" - Day 3 - Advent of Code 2025 https://t.co/uiaDv5nf3X #AdventOfCode

[Matt Florence @mattflo]

I've completed "Gift Shop" - Day 2 - Advent of Code 2025 https://t.co/UGwJwMZOEg #AdventOfCode

[Matt Florence @mattflo]

I've completed "Secret Entrance" - Day 1 - Advent of Code 2025 https://t.co/eur7Kygy7i #AdventOfCode

[Matt Florence @mattflo]

Chunk windowing is an incredibly simple context engineering technique that can improve the quality of augmented replies while keeping your costs predictable.

https://t.co/gSB7ibxWeA

[Matt Florence @mattflo]

I don't tend to re-use prompts often, but here's one that's handy:

Take a look through the changes on this branch. Summarize them. Highlight any issues.

`git diff origin/main...HEAD`

[Matt Florence @mattflo]

The key to healthy code is the same for our bodies -> exercise. How early and how often do you run your code?

[Matt Florence @mattflo]

I thoroughly enjoyed this long form podcast with a legend in the industry, Rich Sutton.

> Large language models are trying to get by without having a goal or a sense of better or worse.

https://t.co/TKJOXU5atX

[jason @jxnlco]

@ebadgio_ @ExtendHQ oh by the way, if you want to learn how to systematically improve your rag applications check out my free 6 email crash course

https://t.co/foNt7kIPCa

[Matt Florence @mattflo]

Finally tried Gemma 3. Impressive all around and really happy with its vision capabilities. This is the 12b model running locally with @ollama and @LibreChatAI. I'm getting a zippy 32 tokens/s on my M3 Max. Who else is going to the @aiDotEngineer World's fair next week?

[Matt Florence @mattflo]

The best agent tool is just an agent with all its autonomy removed.