softwaredoug

These are preexisting trade issues that were being negotiated. The new 50% tariffs are on other industries. Lumber trade issues predate this administration, and aren’t involved in the new 50% that’s put on $20B on other industries.

https://en.wikipedia.org/wiki/Canada%E2%80%93United_States_s...

Lumber is not part of these tariffs. These are on a narrow set of products.

Lumber is a different long standing trade dispute going back to the 80s unrelated to the industries targeted by the 50% tariffs.

Trump created a 10% Lumber tariff in Sept 2025, but it was already tariffed before that. Canadian lumber today isn’t suddenly 50% more expensive.

https://en.wikipedia.org/wiki/Canada%E2%80%93United_States_s...

This is true somewhat, but keep in mind the 50% tariffs aren’t on all goods. They’re on like 5% of exports.

Plus when one side is an unreliable negotiator (the US) there’s not much rational behind continuing.

Exactly. I support remote work, but purely as a question of science, it may not be a direct causal link between remote work -> happiness. There may be other factors at play.

This seems hard to prove a direct causal link.

Remote employees are also in demand enough to negotiate / find a role with remote work. They probably then are experienced, senior, and better compensated than their peers.

It could be that the underlying cause is being more in-demand in a position to negotiate remote work, not remote work by itself.

Where am I wrong?

Exactly

I had a similar experience in search and found even holdouts can be overfit to. IE through brute force, it may not see the holdout, but if you gate a change on holdout acceptance it will land on a solution that’s overfit to it by somewhat random chance.

The other problem is that holdouts / data inaccessible to the agent isn’t easy to do in most coding agents. It’s not as simple as splitting training data 80% and giving some to the agent and hiding 20%. The agent can figure out where its data came from and find ways to reconstruct / cheat the holdout data.

All the ways of doing this seem annoying: ie having a second project that accepts / rejects changes.

I opted to just build my own harness for these things to avoid overfitting.

https://softwaredoug.com/blog/2026/05/17/autoresearching-a-b...

That 33%-39% goes back into stimulating the global economic system where America is at the center. To people like me that own US Treasuries.

You’re not wrong. But it’s not as simple as 33-39% disappearing into a black hoe.

I gave up "keeping up" years ago. I just follow my curiosity mostly.

If something is truly transformational in technology, you'll be reminded of it a lot, and eventually you'll find time to learn about it.

Yes absolutely that's another good trick.

Even better is to search the corpus first with like naive BM25 / embedding search, aggregate over top N to get most representative categories, then have the LLM categorize in that set.

Yes what you're describing is a classic way of doing query understanding.

I've found, though, getting it in the language of the vocabulary has generally improved performance.

Further, when searching for "blue shoes" you want to separate the color from the item type. So its useful to have a dumb LLM do this for you. And with the LLM in the loop, its further useful to get it into the language of the taxonomy to improve embedding retrieval accuracy.

There are of course many ways to skin the cat here :)

Using a Nano model, a tad worse than shipping a vocabulary to a larger OpenAI model. (And it’s an huge improvement on not classifying the queries at all).

But no classification is perfect. In search in particular, you will also want to have places for manual intervention for high priority queries.

I use OpenCode and I like knowing the direct token spend for doing tasks. A healthy repo can get a lot done with Luna + fresh context. Then I can spend $1-$2 a day when I'm doing development, and costwise honestly it beats a $200 / month plan.

I also just do a bit of hand-coding to guide the agent still.

I worry the $200 / month plans are loss-leaders encouraging you to maximize token usage to churn out slop, rather than thoughtfully use coding agents in a way that still engages your brain, and produces good software.

Beyond Meat is one such product, and it generates a lot fewer greenhouse emissions.

> Based on a comparative assessment of the current Beyond Burger production system with the 2017 beef LCA by Thoma et al, the Beyond Burger generates 90% less greenhouse gas emissions, requires 46% less energy, has >99% less impact on water scarcity and 93% less impact on land use than a ¼ pound of U.S. beef.

https://css.umich.edu/publications/research-publications/bey...

Why do people take this stuff so personally?

Eat meat. Don't eat meat. Do whatever you want.

It's somewhat revealing to me that whenever this comes up people feel personally attacked by people trying to make ethical choices using data. Why is it so threatening to you?

IMO we should share data and let people make their own choices.

A couple of decades of not eating meat is a pretty significant improvement on you your carbon impact :)

I would say it’s important to emphasize people don’t need to go complete vegan. Reducing consumption of animal products can be more realistic for most people.

I don't know. Every big tech company I worked at has a lot of turnover. Then every Blind post says "its so over" or "everything is going to change" But that doesn't happen.

It's hard to read anything into it.

Thanks for doing this benchmarking Dylan. I wanted to teach people PCA in my original article, but had no idea it would stack up this well against Matroyshka!

Feels like a “just use logistic regression” moment :)

If I ran a dev team, I’d spend a day a week on deliberately hand coding. Even if the clankers write most of the code, IMO we need deliberate practice to develop taste. There’s a side of taste that’s Rick Rubin, how the customer experiences something. There’s also what a classically trained cellist hears in a more tactile, layered sense in how another cellist plays a piece.

There’s not going to be much of a price difference from Luna (Luna might be a bit cheaper too)

And Luna has actually been my favorite coding model that predictably just works at boring dev tasks.

To be clear, this isn’t wind power currently under construction. It’s proposed future wind power with leases being planned. Trump has largely failed to stop the nearly complete projects in New England, New York, and Virginia.

Still short sighted and stupid, of course.

Usually in these cases you don't need to do much tuning to the retriever. So you just give it BM25 or somesuch.

I'm hesitant to say absolutely zero tuning, because there are cases where you do want to say, bias towards trustworthy results or recent results etc to help the model avoid wasting tokens. But probably not much beyond that.

You can also just create a param in the tool for the agent that selects for "recent" or "popular" or "trustworthy" in ranking.

*MixedBread

https://www.mixedbread.com/

People are building agentic search one of three ways:

1. Actually good retireval. There’s been a lot of progress on serving the kinds of queries agents tend to serve, from places like Hornet, MoxedBread, LightOn. Particularly in late interaction

2. Smarter harnesses with models/judges validating the result. This is now just seen as the generator/ evaluator pattern. Here’s where people try to just use grep or some other naive retrieval system. Let the agent figure it out. But it’ll consume a lot of tokens to get good results as it iterates and loops.

3. A model trained for retrieval. Give it dumb retriever like in (2) but it is fine tuned on the task as in (1).

This article is 3. But we’ve been seeing this all year with SID.ai, Gleans Waldo model etc. if this interests you I’d check those out, particularly SID.

I wrote about these 3 approaches here https://softwaredoug.com/blog/2026/06/08/three-kinds-of-agen...

What I really mean is a difficult to revoke license. Not dependent on a server or DRM. I realize I do not actually own copies.

I can buy and own legal copies of movies and music. But I often really can’t with games.

There’s also a class of “detailed agent written technical blog” that’s you can just scan and get the shape that it’s long paragraphs, tables of statistics, and dry LLM writing. You notice the blog has one or more such posts a day.

It’s a turnoff immediately if someone shares what their agent wrote.

It also has happened a few times they contain a fair number of misrepresentations and mistakes.

Recently got a big fancy TV with 4k + Dolby Vision

It’s a lot of work figuring out what movies support the format. And all the premium upgrades to Netflix, etc get them to stream premium formats. Then it doesn’t work and you’re spending 30 minutes debugging your TV, wondering why you’re not getting the amazing new restoration of 2001: A Space Odyssey.

I totally understand why people just buy physical media.

Games may be harder given software updates. But I’d pay a premium for a physical copy of mature games I love (like Halo Master Chief collection)

I don’t think this is true.

Opposition is not really related to aesthetics.

People fight even theoretical housing (zoning). NIMBYs oppose building housing on derelict / ugly lots. Architectural review is used by neighbors to stop housing - not because they’re preservationists - but for a load of unrelated issues and architecture is a convenient veto point.