Michael Bleigh

I hate when I think about a problem for a really long time but can't navigate the tradeoff space to a clear ideal outcome. It's not fun to defend a position that feels "least bad" instead of "good".

There needs to be a benchmark that mixes common acronyms with a few undefined fictional ones to test whether the model hallucinates the acronym's expanded definition. I see this even on frontier models, even when they have the tools available to search for and define the acronym properly.

In Pluribus, an alien intelligence usurps overnight and dissolves all previous divisions, establishing a semi-utopic society with regretted casualties along the way and a small group who refuse to partake of the new order. So it's about AI and the software industry, basically.

I find myself frequently telling LLMs to explain things like I'm a "semi-technical exec". It's actually not a bad way to get more comprehensible technical writing out of them. 😂

"This in-app agent could have been a skill" is the new "this meeting could have been an email". If people at your company want to build agent functionality that can only be used inside your app/product, tap the sign:

Unpopular opinion: plan mode is still good and correct. Now that models are more powerful and do longer horizon work it's more important to be aligned up front. A dedicated plan artifact is easier to read and review than inline chat content.

I would estimate a high percentage of all software (80%+) would be totally fine with only automated reviews. Because 80%+ of software written is relatively low risk or changes are easily reversible. AI can free us to focus on the changes that actually do need expert attention.

It used to be that incomplete software was simply missing many things that needed building. Now incomplete software has all the pieces, but the pieces are bad (i.e. slop). Finishing a project is becomes more about removing bad ideas and polishing good ones. This isn't good or bad, just different.

I'm seeing SO MUCH Jev in my feed. My theory is this: most people don't want to train models, they want to build things. So just like how LLMs broke through to the mainstream because they didn't need to be trained to be useful, such is true for Jev and classification tasks.

When AI builds an interface.

It's amazing how frequently "AI did this really stupid thing" turns out to be "I gave AI a poorly-specified prompt and it made different assumptions than I expected". Current models are (usually, not always) good at following instructions if you give them good ones to follow.

Hot take: trying to be more careful about merging AI code is the wrong long-term approach. The better approach is to eliminate one-way doors and make most code more disposable. Sloppy code can be polished/replaced so long as you keep reins on API surface and storage schema.

I genuinely think the demand for tokens is effectively infinite and is bounded only by cost constraints. We will build legions of automatons that are experimenting, tweaking, optimizing, reviewing, condensing, and making sense of the universe of information 24x7x365.

AI code is like water: it will slosh around to fill whatever container it's poured into. If you don't provide any constraints you're going to end up with a big mess on the floor. The new job of software engineering is to specify verification+guardrails that hold the water.

What we need as a companion to WebMCP is WebACP that lets me bring an agent of my choosing into the browser. I don't need my agent to build a browser and I don't need my browser to build an agent. I need my agent in my browser.

Doing crazy esoteric stuff in Go: 3 stdlib calls Doing dead basic stuff in Rust: 3 crate installs

As companies grow, they build processes that leave no room for judgment calls. The default posture becomes oriented around fear of idiot / bad-faith actors, putting a huge tax on good-faith actors who have a good reason to do something a little bit risky. It destroys velocity.

Agents' overeager coding makes us all into editors. The PR is ready not when there's nothing left to add, but when there's nothing left to take away.

Not supporting Cmd/Ctrl+Click to open new tabs on navigation is the usability papercut that most frequently sends me into a state of barely-contained rage when using an app.

Bigcos love to add approvals to everything without considering desensitization. Any individual has limited "pay close attention to thing I'm approving" bandwidth. The more often approval is required, the less carefully each review is performed. Rubber stamps become the norm.

If you *actually* care about the security of your platform you have to also care about usability. If it takes 35 steps for users to do the right thing and 3 to do the wrong thing, you are incentivizing bad practices. The "pit of success" is hardest to construct for security.

How has Apple been allowed to get away with the absolutely dismal quality of file search on Mac? It literally couldn't find a folder called "records" in my Downloads folder when I searched for "records".

One pattern I find myself reaching for frequently these days is to gather data into a heavy JSON "kitchen sink included" format and then build lightweight CLI renderers that slice and dice the data into various token-efficient text views for an agent to understand.