David Whitney

Gotta say the explosion of decomp/recomps has made me pretty nervous about not angering the gods of Rockstar with my GTA1 Remaster that uses similar techniques. Will ruminate - it's wonderful, has a fixed / improved camera, uncapped framerate, modern display and multiplayer and gamepad support.

That's the frustration between the "vibe coded slop" and the carefully crafted work. Both can involve AI, but the two things are really not the same genre of thing.

Frequently work with teams who are uncomfortable changing software they own because of low test confidence and / or coverage. Generated this small tool for .NET projects (it's written in TypeScript, don't loose your minds) which acts as a meta-harness, instruments a codebase with Coverlet + Stryker

If we're now in a world where you want to do risky changes with agents or big traditional refactors, starting from a place of confidence is important, give it a go. github.com/davidwhitney...

Frequently work with teams who are uncomfortable changing software they own because of low test confidence and / or coverage. Generated this small tool for .NET projects (it's written in TypeScript, don't loose your minds) which acts as a meta-harness, instruments a codebase with Coverlet + Stryker

There's a strong focus on purely outside-in tests, rather than unit testing, because anything more granular would lock you into your specific implementation and make your code harder to change.

Frequently work with teams who are uncomfortable changing software they own because of low test confidence and / or coverage. Generated this small tool for .NET projects (it's written in TypeScript, don't loose your minds) which acts as a meta-harness, instruments a codebase with Coverlet + Stryker

It has canned prompts that do their best to mimic your existing test style and only isolate out of process dependencies, with a deterministic shell that measures coverage and mutation test rules, letting the agents only do the test uplift, rather than an agent judging the work.

Frequently work with teams who are uncomfortable changing software they own because of low test confidence and / or coverage. Generated this small tool for .NET projects (it's written in TypeScript, don't loose your minds) which acts as a meta-harness, instruments a codebase with Coverlet + Stryker

I've built and rebuilt this "thing" over the past 2.5 years ~4 times. It's finally taking real shape as a thing that other people will be able to use. A code / entire platform inspection tool - that builds code graphs and supports agents with context via MCP. "What if SonarQube was useful?"

One of the important differences (to me) about the approach I'm taking with this is it's zero-config by default - you don't wire it into your build pipelines, instead you either connect it to GitHub or point it at a local set of repos and it introspects and builds graphs there.

I've built and rebuilt this "thing" over the past 2.5 years ~4 times. It's finally taking real shape as a thing that other people will be able to use. A code / entire platform inspection tool - that builds code graphs and supports agents with context via MCP. "What if SonarQube was useful?"

I know it might seem like an losing battle to care about how code is composed in an increasingly agentic software dev landscape, but I still believe: -Code is load bearing, better code leads to better systems -The models require structured language to steer true, and that is encoded in quality code

Every time I see someone post this, I wonder if they think the start menu is legitimately opening a WebView? React Native != React.

Sorry, wait, let me write that in JavaScript real quick React Native !== React

Agree with a lot of what @robbowley.net is saying here. Then current news cycle about AI safety and slowdown is bullshit - market fixing disguised as regulation. We need two things: - Models trained on public data legislated into the open - Governance as architecture, not a policy and process

Example: - Agent hosts model - Model uses tool that access customer data - Models internet access tools are disabled and the sandbox is marked as "tainted" to prevent exfiltration - Agent cab continue interacting with customer data safet This is the future of models and agent governance imo.

Agree with a lot of what @robbowley.net is saying here. Then current news cycle about AI safety and slowdown is bullshit - market fixing disguised as regulation. We need two things: - Models trained on public data legislated into the open - Governance as architecture, not a policy and process

Small hobby project off the side of a hobby project. I'm doing some game preservation/restoration on GTA1. 32/64 bit support, Mac/Linux port, modernized controls and camera. Here's a native Mac port...

The internet has been living through the Eternal September since 1993 and every tech shift exposes the then-current generation of users to that feeling that something intrinsically of value, that only existed digitally, has been lost. It's weird to have lived through many of these Septembers.