Michael Gwilt

@davidfowl @sinclairinat0r @JamesNK Seems useful for helping an agent find which PR caused a regression. More descriptive than commits, less verbose than diffs

. @JamesNK built an agentic workflow to generate a change log for aspire github.com/microsoft/aspi… - We have 400+ PRs in the next release 🤯 - The workflow categorizes fatures, bug fixes, areas of the project etc. - It's incremental and processes unprocessed PRs on an interval

@aj_dev_smith Hell yeah!

dev log day 3. today's all about lighting

@aj_dev_smith That’s gonna look sick with the rest of your lighting if you can get it to work well

@mgwilt thanks! and yeah, don't see why not as they're procedurally generated... 1 prompt later and here we are! need to make some tweaks, but it already makes a big difference! I'm 100% trying this in darker scene later.

dev log day 2. sound on 🔉

@davidfowl @aspiredotdev 👀

Yea, this one is a banger. #aspiredev @aspiredotdev

@davidfowl @techgirl1908 I combat this with strict adherence to well known design patterns. Where before I was a little more relaxed, today code has a correct and incorrect place to exist. The only way to avoid duplicate code is to make it so the LLM will always try to put the same code in the same spot.

There's a really high cost of duplicating code and we're all relearning that everyday with these coding agents. Until they get better at generating reusable code, bugs will run rampant. You can't "abstract it away" with more agents. You just get crappy software....

@davidfowl yes.

Are you spending your tokens on breadth or depth?

@sama The compaction and code quality differences between Codex and competitors is night and day. For shipmas, all I want is a mobile app that tunnels to a codex cli or the app, though 🙏

It is very nice to see Codex getting so much love. We are launching a $100 ChatGPT Pro tier by very popular demand.

my yolo11 fine tune summarized

@davidfowl I was using Aspire + Dapr for a chat app experiment and it was wonderful. Being able to run both python and dotnet services along with all of the arbitrary supporting infrastructure like redis/postgres with F5 was a dream.

Building this chat app is fun just to dogfood and play with new features. Can't get over how helpful this view is for developing applications that talk to literally anything. #dotnet #aspire

duckdb might be the most useful tool in my dev arsenal

No dedicated devops. Rotating responsibility. A black box can’t be optimized or innovated on. Enlightened developers make a point to be well versed in every aspect of the system they work on. Particularly in the cloud where the choices you make on how you deploy your application directly impacts the architecture. Makes me sad to see developers toss things over the fence.

Thoughts on dedicated DevOps teams vs DevOps role within the team vs software engineers owning their own DevOps?

Editors should support rendering mermaid in comments

@theo Turns out that LLMs are really good at writing prompts too. My favorite workflow is having an LLM help me write a detailed spec and then making a whole new context to work through the spec one section at a time.

Hot take: typing speed matters for code now more than ever. Some of these prompts are long as hell...

Do they have insurance companies in the Netherlands

@sama deep research is awesome. even more awesome would be if I could point it at an internal knowledge base if I exposed an MCP server to it.

@ThePrimeagen Have you tried just asking it nicely

i need help on prompts.... copilot like prompt, wtf. My model just now wants to mansplain everything what is your best attempt at getting a model (qwen 2.5 coder 32b q5) to respond with just one line of code

indexing 8gb of documentation. today ain’t gonna be cheap

oooo cursor added support for MCP servers… that’s a huge deal.

@theo If the link doesn’t change or expire, then it’s not a hallucination, it’s just in the training data.

v0 just hallucinated the CORRECT LINK TO JOIN MY DISCORD WTF