Matt Makai | Full Stack Python | Plushcap

@lizziepika 👀 cool thanks, makes sense!

@fullstackpython Yes! Runners are configured per-agent + per-model, so you can def point diff runners at different models on the same change (1 on GPT-5.x, another on a different model...each producing its own findings/scores.) I'm not sure models are wired up by default tho

He's right! The PR and reviewing code is a dying paradigm. New feature we're testing @ Entire: agents review every change + file findings, ranking by severity, tracking to resolution, and sometimes automatically fixing. Then, gates decide what merges. Risk too high? Blocked. Still want a human in the loop? "Peer approval required" is just another gate. Devs define what "ready to ship" means, then Entire enforces it.

@lizziepika 🙏

@fullstackpython so helpful!

Several articles & resources I’ve been reading lately to learn about how LLM architectures have evolved: * finbarr.ca/five-years-of-… 5 years of GPT-class model progress, only goes to 2023 but that puts more focus on early models where details are precisely known * magazine.sebastianraschka.com/p/from-gpt-2-t… great companion post to the first one as its diagrams and code examples are absurdly helpful. @rasbt has so many articles that could be on this list but I picked this one because of the visuals in particular * gregorygundersen.com/blog/2025/10/0… denser than other posts. my attention can glaze over some of the deeper math formulas but as a synthesis of key LLM concepts have evolved this is excellent * cameronrwolfe.substack.com/p/the-history-… both accessible in the basics "what is a token?" and detailed, this one goes into other notable models from early LLM days (post image is from this article)

@heccbrent nice! thanks Brent

@fullstackpython Yes! 5.6 is really good at inspecting frames to find something you want it to look for. I've attached a recent example from my work. Prompt idea: "find 3-5 clear stable frames where the presenter is smiling or making a bold expression"

Sorting through video footage from a shoot is tedious. It doesn't have to be. Give Codex a folder of A camera, B camera, audio files and (optionally) a script/transcript and ask it to identify the best takes!

@thdxr Built Plushcap for dev trends data so you know what’s real (harness engineering) from flash in the pan (OpenClaw, tokenmaxxing). Hand coded started in 2022, CC in 2025, Codex since March 2026. Made a lot more progress since using CC+Codex plushcap.com

please i'm begging you show me something you built not another "this is my custom agent setup" post where you pretend you're doing something smarter than vanilla claude code please

With 5.6-Sol and Codex, this is a workplace. I'm sitting poolside dictating updates to the model via my phone. It edits a video in DaVinci Resolve and uploads the render to Google Drive so I can review it from my phone and iterate again.

@heccbrent 😯😯😯

@lizziepika @EntireHQ 😯😯

Git hosting was built for humans slowly click-clacking. @EntireHQ is rebuilding an *Entirely* new Git hosting network for the agentic era. Git started by mirroring your repo today: entire.io/blog/an-entire…

Daily reminder what agentic harness is

@clairevo Makes sense! Longer uninterrupted runs = a lot more tokens. Thanks Claire

@fullstackpython I think it can just be more autonomous and long running so can do sprawling tasks in a denser amount of time. I'm consistently running it overnight now, which I wasn't doing before.

my current fable API cost run rate for a single engineer (me) is about $200k a year jfyi

@crtr0 yea agree, I feel that personally too

@fullstackpython Not as much, and I think AI is a big reason. Developer comps are moving up the stack, so it's more about frameworks & platforms. This is a good thing ™️

It's absurd how fast you can improve a web app's agent readiness, performance, SEO/GEO, security, accessibility, and more by pointing Claude Code or Codex at this Website Spec site (specification.website) with a prompt like "Check the current implementation against this set of agent readiness specifications and report on what the application does well, where it needs improvement, and what you recommend we fix immediately: specification.website/spec/agent-rea…" Run that prompt while in your codebase. You don't need to fix everything all at once, but you're guaranteed to make huge improvements in all of those areas just by using that specification site to incrementally improve with an AI coding agent. If you want to see an example of what's generated before trying this yourself, take a look at this example output I had from running it on Plushcap's live website and codebase: plushcap.com/blog/improving…

@bnivanovX 🙏 still early days for harness engineering it’s great to see such high quality resources already exist for it

@fullstackpython Brilliant sources, thanks for sharing!!

A few resources on harness engineering I'm really enjoying learning from recently: * walkinglabs.github.io/learn-harness-… - excellent course that covers how to constrain agent behavior in a productive way, what the challenges are using LLMs especially maintaining context, and how to make the harness more agentic and work to the proper conclusion of a task * twotimespi.dev - build a simplified agent in Python (Tau). start with "uv tool install tau-ai" * humanlayer.dev/blog/skill-iss… - lessons and learnings from failing and succeeding to build harnesses for coding agents over 12+ months * langchain.com/blog/the-anato… - higher-level overview of the many components that need to come together properly in a harness to make an agent work

I meant to tag @_alejandroao on this post. Thank you & the @huggingface crew for Tau, such a great idea and implementation!