Wesley Boland

building the thing to build anything. videos ↓

[Wesley Boland @wesleyboland]

building an app is becoming one prompt. knowing it actually works isn't.

an agent's effectiveness is proportional to its ability to verify its work.

no tests it can run itself? you get a first draft.

good, better, best verification for claude code & codex 👇

https://t.co/wvJM9oKcFo

[Wesley Boland @wesleyboland]

@poteto chat is too slow. in the near time, i want more agent hosts to adopt mcp ui w/ elicitation https://t.co/XdrosxaToy

[Wesley Boland @wesleyboland]

the old bet: the value is in the harness, the model just needs to be competent. the new reality: models are eating the harness. subagents, planning, tool use, all built in. does the model always win?

[Wesley Boland @wesleyboland]

@SamVale29 ...what do angels do in their spare time?

[Wesley Boland @wesleyboland]

@poteto chat is too slow. in the near time, i want more agent hosts to adopt mcp ui w/ elicitation https://t.co/XdrosxaToy

[Wesley Boland @wesleyboland]

agent verification is all you need and it’s analogous to TDD. Use a process that constraints and shapes the output for agent to maximize https://t.co/oAG14vBUiy

[Wesley Boland @wesleyboland]

@addyosmani the ability for agents to verify feels like an outcome of good architecture. are folks starting with in mind? https://t.co/sxeXrK1v0g

[Wesley Boland @wesleyboland]

claude & codex before they say "done": check yo self before you wreck yo self

https://t.co/Y5v6kWy981

[Wesley Boland @wesleyboland]

building an app is becoming one prompt. knowing it actually works isn't.

an agent's effectiveness is proportional to its ability to verify its work.

no tests it can run itself? you get a first draft.

good, better, best verification for claude code & codex 👇

https://t.co/wvJM9oKcFo

[Aaron Ogle @geekgonecrazy]

@ibuildthecloud @wesleyboland This looks a lot like what we’ve built at atomic with intents. Cool part is they are then linked together with the change in the semantic change graph.

[Andrew Haigh @a_haigh_]

@shadcn You really think it's the model not the harness/ process?

[kache @yacineMTB]

good software engineers are graduating from managing codebases to simply just managing complexity

[Wesley Boland @wesleyboland]

@carlassmann everything is in the codebase, including the software factory. agents should be able change the factory too.

[Wesley Boland @wesleyboland]

for example, datadog/sentry’s internal metrics must scale higher than the app its monitoring (sampling be damned). for a software factory to effective it must understand the meta properties of the product being built - does seo matter? does compile time matter? is the product gaining traction? are stories/issues being stuck in some dev state?

[Wesley Boland @wesleyboland]

@gmkurtzer …when the LLM says how to.

[Wesley Boland @wesleyboland]

the analytics tell me what to do next, but the direction feels ‘copy what everyone else is doing’ and less ‘be unique and special’. i guess the ‘work’ begins...

[Wesley Boland @wesleyboland]

@dark_coderz agent native architecture first, microservice last!

[Wesley Boland @wesleyboland]

@coryalthoff the work was always there, we were just comfortable

[Wesley Boland @wesleyboland]

@ibuildthecloud did that, but it was too meta for most to appreciate. we’re still in the skeuomorphic phase of factories. https://t.co/cHQba4Y88t