iiinova — where your digital self keeps evolving

Michael Gwilt @mike.dev · 21/10/2025

What’s more concerning is that anybody would want a *heated* bed

260

Michael Gwilt @mike.dev · 21/10/2025

LLMs can do incredible things. This ain’t it though.

000

Michael Gwilt @mike.dev · 21/10/2025

I just had to talk to a chatbot at a doctors office and I think I developed a heart condition in frustration. They replaced the normal intake form with a thousand chatbot turns.

120

Michael Gwilt @mike.dev · 21/09/2025

Everyone is all hyped about LLMs and here I just want SAM 3

000

Michael Gwilt @mike.dev · 10/09/2025

The only thing I don’t like about MCP is that clients make me beg to include tools in the plan. Like, just give me a checkbox to say you MUST schedule a tool call in the plan.

000

Michael Gwilt @mike.dev · 08/09/2025

I have! But I don’t have the patience for spec driven development most of the time, even though I know it would help.

010

Michael Gwilt @mike.dev · 07/09/2025

AWS has to contend with Google absolutely killing it with dev tools and libraries for AI workloads. AWS has traditionally struggled with making tools that developers actually want to use. AWS might pull ahead in because of market share, but I’ll be surprised if they come out ahead in capabilities.

120

Michael Gwilt @mike.dev · 07/09/2025

I feel like a lot of teams start loose and fast with Firestore and then find themselves in a pickle as soon as they start growing a team. What fields does this collection have? Who knows, better iterate over every single item just to figure that out ☠️ No, the only answer is a strict contract

000

Michael Gwilt @mike.dev · 07/09/2025

I can already imagine a persons whole job being chasing down teams trying to standardize field names

100

Michael Gwilt @mike.dev · 07/09/2025

I literally cannot work with Firestore without seed data in json. I lose track of fields almost immediately, otherwise. How does Firestore scale to large teams? Do you end up with a ton of spaghetti checking if slight variations of field names exist?

100

Michael Gwilt @mike.dev · 29/08/2025

When the commits devolve to literal begging for a green check mark

010

Michael Gwilt @mike.dev · 29/08/2025

Had a dream that a rust app was silently crashing every time I tested it. That’s it. That was the dream. Over and over.

000

Michael Gwilt @mike.dev · 26/08/2025

September 8th, the good PNW weather begins. I desire cold

000

Michael Gwilt @mike.dev · 21/08/2025

I think the response to 4o being shut down alarmed people. It put a spotlight on the psychological impact chatbots have on some folks.

040

Michael Gwilt @mike.dev · 17/08/2025

These things are more relatable every day

040

Michael Gwilt @mike.dev · 16/08/2025

To suggest you can’t learn from something just because it’s not always 100% correct demonstrates an unfortunate character flaw. I learn from my peers, but I don’t go expecting them to be know everything. Part of the learning process is discovering the correct answer together through trial and error

080

Michael Gwilt @mike.dev · 16/08/2025

segoe ui bold, if you will

020

Michael Gwilt @mike.dev · 16/08/2025

Windows Phone UI was inspired. And the colors of the Lumia series were amazing. I used to get so many comments on my cyan phone. If Microsoft tries another handheld device, I hope they’re bold on UX again.

1100

Michael Gwilt @mike.dev · 15/08/2025

.NET Aspire makes working with Dapr 1000% more comfortable

073

Michael Gwilt @mike.dev · 12/08/2025

Not perfect but it kind of works. Speaker detection of multi-user webcam input.

010

Michael Gwilt @mike.dev · 12/08/2025

My favorite thing about wandering into Applied AI as a focus is that in 200 lines of code you can make something profoundly useful. Like, the dopamine efficiency is off the charts, especially in CV.

000

Michael Gwilt @mike.dev · 09/08/2025

Trying out Dapr Agents today. I tried Dapr+Aspire about a year ago for this purpose but they didn’t have a dedicated concept of an agent.

000

Michael Gwilt @mike.dev · 08/08/2025

“I ain’t reading all that” as a service is actually useful though.

500

Michael Gwilt @mike.dev · 07/08/2025

Oh and there’s this docs.litellm.ai/docs/provide... which has a sample for tool calling

docs.litellm.ai

Ollama | liteLLM

LiteLLM supports all models from Ollama

000

Michael Gwilt @mike.dev · 07/08/2025

Another way would be MCP servers. Anthropic has a platform agnostic spec for tools that you can check out here. github.com/modelcontext... They have an SDK also that lets you interface with these OSS servers that facilitate things like browsing or any domain specific task.

github.com

GitHub - modelcontextprotocol/servers: Model Context Protocol Servers

Model Context Protocol Servers. Contribute to modelcontextprotocol/servers development by creating an account on GitHub.

100

Michael Gwilt @mike.dev · 07/08/2025

It’s kind of jumping into the deep end, but I think the full stack sample from this tool is great. github.com/GoogleCloudP.... It’ll replicate DeepResearch using the Google Search tool. The ADK docs show how to use LiteLLM for agents which is an abstraction for basically any inference provider

github.com

GitHub - GoogleCloudPlatform/agent-starter-pack: A collection of production-ready Generative AI Agent templates built for Google Cloud. It accelerates development by providing a holistic, production-r...

A collection of production-ready Generative AI Agent templates built for Google Cloud. It accelerates development by providing a holistic, production-ready solution, addressing common challenges (D...

100

Michael Gwilt @mike.dev · 07/08/2025

Stay skeptical. They’re very useful, but it’s almost as if as soon as you put too much trust in them, they demonstrate their limitations. Very fun tool to try to keep up with though. It’s all moving very fast.

010

Michael Gwilt @mike.dev · 07/08/2025

Try pairing it with a tool that looks at the documentation itself. I wouldn’t trust a super quantized model to not have important data missing. If the information is in the same context, it might not mess it up. Sometimes I literally copy and paste entire pages of docs into my prompt.

140

Michael Gwilt @mike.dev · 04/08/2025

I still think Dapr is one of most under utilized frameworks for agentic app development.

000

Michael Gwilt @mike.dev · 01/08/2025

if docker compose supports llm endpoints, that means I can start models in my devcontainer definition.

000

Michael Gwilt @mike.dev · 24/07/2025

so the next big phone tech competition is certainly focusing on AI chips right? now that I’ve experienced a high quality model in my palm, I want m o a r

010

Michael Gwilt @mike.dev · 23/07/2025

Got gemma-3n working in an expo app today on an android device. It’s pretty quick, but boy does the phone get hot quick.

000

Michael Gwilt @mike.dev · 26/06/2025

@unrealengine.bsky.social Would be wonderful if you gave your documentation some love in this same vein. While extensive, it’s difficult to find exactly what you need especially if you don’t know the exact thing you’re looking for. An LLM would help here.

000

Michael Gwilt @mike.dev · 26/06/2025

OpenAI’s Deep Research is very interesting when you tell it to limit its research to a single website. In this case, I limited it to app.leg.wa.gov and it just ripped through Washington renters code and found exactly what I needed along with supporting information from other sections.

app.leg.wa.gov

100

Michael Gwilt @mike.dev · 25/05/2025

Owning physical copies of books is awesome until it’s time to move.

000

Michael Gwilt @mike.dev · 02/05/2025

I don’t see the problem :)

150

Michael Gwilt @mike.dev · 23/04/2025

Having been there, I sympathize. It makes you feel profoundly useless when you see your team struggling on a deadline and you have the technical background to help but can’t because you’d get in the way. But its important to remember that if you haven’t been hands on the whole time don’t start now

130

Michael Gwilt @mike.dev · 13/04/2025

Spent the day digging into supabase and expo. There’s so much cool stuff I was missing in big enterprise fintech dev

010

Michael Gwilt @mike.dev · 08/04/2025

🧀 the gold standard 🧀

000

Michael Gwilt @mike.dev · 04/04/2025

I’d really like to see Microsoft.Extensions.AI.ModelContexProtocol The repo that Anthropic picked up is pretty ok but I think this is a gap that Microsoft could fill in a more robust way in-framework, especially since they’re integrating the spec into their products.

000

Michael Gwilt @mike.dev · 03/04/2025

I appreciate how re:Invent throws frugality to the wind and prioritizes the customer in an almost obscene way.

000

Michael Gwilt @mike.dev · 03/04/2025

Ultimate goal is to make a C++ implementation of MCP, but for now it’s easier to use anthropics incubating rust-mcp repo.

000

Michael Gwilt @mike.dev · 03/04/2025

Converted my Unreal Engine MCP server plugin from Python to Rust and cut out having to have a TCP server from UE->Python Now it’s Unreal C++ <-> Rust through C bindings. No more Python dependencies, just a dll to drop into your binaries folder The majority of the code is C++ anyway

110

Michael Gwilt @mike.dev · 03/04/2025

Will it shrink to fit the icons better if we disable the search and weather widgets?

010

Michael Gwilt @mike.dev · 03/04/2025

I’ve had difficulty getting some folks to use containers. Some developers just want to code and not learn anything that looks like infrastructure, which I understand and can respect. Aspire is an opportunity to ease people into the world of containerization.

media.tenor.com

a baby eating a slice of pizza from a blue plate

ALT: a baby eating a slice of pizza from a blue plate

000

Michael Gwilt @mike.dev · 31/03/2025

A small experiment today. MCP server wired up to Unreal Engine

000