It is actually pretty simple unless you are a bot, of course. The vikings didn't annex russia, they have merely replaced the rulers with one of theirs, keeping the rest as it was.
The mongol conquest also hasn't been an annexation. It was more like a racket, with the russian princes paying protection money while being free to launch wars of conquest on their own. Pretty common in history, really. See danegeld as an another example.

How do you decide to trust code you write?

How do you decide to trust code someone else wrote?

Tests. Design discussions. Code reviews. Code review questions. Design Questions. "What about this problem?" Real validation testing.

I was in a funny situation last week. Two different humans told their claude agents to fix the same bug. Claude came up with two _slightly_ different implementations.

I told _my_ claude agent to code review both implementations - using a new prompt I'm working on to try and automate away a lot of the code review value I was adding manually.

Same model; same bug, two implementations, and the code-reviewer agent found different bugs in both PRs.

We ended up blending the best things from both PRs into the one that shipped.

I am really confident that the code I ended up shipping here is better than what our specific humans would have done without using AI, and 10x-100x faster.

I am spending more of my time telling agents how to think like me, what to look for, what tradeoffs to make, etc. They wont do those things unless I tell them to. They don't always do those things perfectly.

(Neither do the real human software engineers who work for me.)

That is wildly optimistic IME. Frontier models do a great job of sounding authoritative on any given subject. This is not the same as actually having expert-level knowledge and understanding, and the choices the models make when you ask them to implement something can and often do reflect that. There are several domains I've worked in over the past few years where AIs lack even basic knowledge an expert practitioner would have, probably because these fields are too specialised to have a lot of books available or papers published, so presumably there hasn't been much subject-specific content to train the models on.

I don't think it's usually about what AI "did really badly on" at this point. It's more about whether the overheads of adding all the guard rails and then reviewing and correcting all the output end up costing at least a similar amount to just writing the code yourself.

For a junior developer, I'd say frontier models passed their standard some time ago. Today's LLMs routinely write code at the level of a competent junior-mid but with the advantage of an inhumanly broad knowledge base.

For a good senior, I'm much less convinced. The frontier models are hit-and-miss when it comes to writing good test suites, which matters a lot when they rely on tests so heavily to reach acceptable quality. IME, they are still mostly miss when it comes to the business insight that seniors will take into account and design/architecture decisions that keep the code manageable as it evolves. While they have become much better at recognising existing patterns in the code, and I would say they follow them reasonably well the majority of the time now, there are still bizarre anomalies that creep in, and they aren't very good at extending those patterns when necessary.

I think AI advocates also tend to underestimate just how much time it takes to properly review what they are doing. Take a half-day Claude session to implement some new feature. It might take half an hour to give it the initial brief and then review its plan. Make a few edits and it'll propose a new plan that is also long and detailed and has no edit markings to show what it changed so you have to review the whole thing, carefully and in detail, again to pick up the minor changes it also made for no obvious reason. Then you let it start doing its thing. You can't really leave it unattended for long because it'll stop to ask questions. Some of those questions might be whether it can run a screenful of ad-hoc Bash script to make some edits to your source files, because for some reason it decided not to show the actual edits as it usually would, so now you have to figure out what all the sed incantations are actually going to do while looking at those files in another editor or just trust it and then review all the changes properly later anyway. Sometimes it decides to change its plan mid-flight, but not always in a good way, and if you interrupt it to apply corrections or redirect, you break the neat plan flow it was in. Then you get to the end and you can finally see how everything it has done fits together, and you still need to review everything with the same care as you would with a junior developer's code and probably apply a few fixes where it's come up with weird names or not quite followed a pattern or imagined some instructions that were never there and made other changes in other parts of the system that should be out of scope.

If you read to the end of that last paragraph, you know how I typically feel at the end of a half day working with an agentic UI. And the thing is, I do those sessions as experiments fairly often to keep up with progress, but a lot of what I do is still written by hand, which gives me simple, objective data to compare the two ways of working and see whether the AI tools are really making me more productive once you take everything into account.

Currently, based on objective data comparing similar fully completed development tasks like for like, I still outperform agentic workflows using frontier models by about 25-50% on average, and I don't have to pay anyone extra fees to do that, and subjectively I also find it more engaging to work that way. So I don't really buy the hype about AIs making us all orders of magnitude more productive unless you're happy to YOLO everything forever and take the much-discussed risks that comes with.

Personally, I think the optimum use of LLMs right now is probably not to use them for the whole coding workflow like that or rely on agentic tools, but to use them for tasks like answering quick questions about the existing code, doing research like a fancy search engine, writing a first draft of some boilerplate, and writing quick and dirty throwaway code for experiments, one-shot tools and the like, where you don't really care about the quality as long as it gets the job done and you will see immediately if it has in fact got the job done. For these things, the LLM can read and type faster than me, and that's useful.

Use a paid subscription to a frontier model. Make sure it knows how to run your build and run your tests.

If you're worried, have it write tests. Look at the tests; see if they look like the right tests.

Tell it to ask you questions if it needs to resolve something ambiguous. Have it explain what it thinks it should do to the code. Ask it probing questions if you aren't sure it understands. Ask it about design tradeoffs.

I have yet to find a problem where AI generated code doesn't get me at least 90% of the way there. Multi-domain, multi-language, multi-platform. That includes stuff I was doing a year ago. It's gotten dramatically better in 2026.

If someone has an example of a software engineering problem that AI actually did really badly on, I would like to know the model and see the prompts. I'm not saying it doesn't exist. I'm saying I haven't seen it yet, and I'm really curious.

I've done software development professionally since the 1990s. I started coding on the Apple ][ in 5th grade. I've had AI do the wrong thing, or misunderstand something, or miss something that seemed obvious to me. But on all of those same criteria; I've had human software engineers do far worse, far more often.

There are 640 acres in a square mile

[Sighs in metric]

Microsoft has probably the most illustrious history of bugs where just visiting a website or opening an email can get your system hacked, but you're right, this is not a problem unique to them.

These days, Lookout asks you before it loads any images from outside your organization...

I fully admit that this might be a 'me' problem, not a them problem; but I just don't understand why this agent monitoring setup was so vulnerable and is being treated as so novel.

Sure, the monitoring of whatever goes on inside the LLM is comparatively new and exotic; but when you set it loose on the internet it's going out through one or more of your WAN links, presumably with your router and firewall and so on in the way, and is mostly doing HTTP(s) stuff. Keeping an eye on what an internal host is doing when you can insert whatever trusted roots you want for TLS MiTM and you 100% own the pipe that it is using to communicate is not necessarily a desperately enjoyable task; but it would have been a relatively banal requirement a quarter century ago.

I'd also think that, when you are spending $$$$$ on the exercise in Nvidia compute gear, and it's a test of a 'frontier model' whose behavior and capabilities you are trying to better characterize and refine, allocating the handful of terabytes necessary to just store a full packet capture of the session in case it is relevant would be a no-brainer item.

Is there something I'm missing that makes this case more complex? I'm not talking about the 'grubbing around in the LLM's vectors and such' side; just the "a host is doing HTTP stuff through a WAN link we control, can we observe what it is doing?" aspect.

The proper response to being in an area where datacentres are crowding in is "lobby your elected officials to increase property taxes and not give out tax increment financing deals". Tax everything, and tax it well. They'll stop building new datacentres in your area, and the ones that are there will (unless they were given a long-term sweetheart deal beyond your already-low tax rates) start being cash cows for your area.

Hello fellow Frame owner.

I bought a Frame 65 just because of how beautiful the display was in Art Mode, and I want pretty art in my house for the 90% of the time the TV would otherwise be off.

I use the "Art Mode", but only with my own sideloaded content.

One thing that stinks is that sideloaded art doesn't support the full Art mode experience; specifically no Metadata. I seeded the display with lots of public domain images of classical and romantic european art (my taste - and why I never considered a paid art mode subscription), and I'm not cultured enough yet to just remember what all of it is. So I wish I could get the tv to show metadata sidecars for my custom images.

The Samsung firmware is garbage. I hate it whenever I accidentally hit the remote button for voice assist or AI assist or any of that garbage.

But the way the display looks when showing paintings is too good to pass up. If i could get a firmware that took out home screen nonsense and disabled 80% of what the TV did, but added metadata display for sideloaded images... it would be perfect.

If you see that, they are usually friends.
Knowing each other long time.

I saw a raven stealing food from a bald eagle. Pulling his tail feathers, the eagle did not even get angry. He just stepped back after pull #3 or #4 and let the Raven eat the rest of the fish.

I always keep this in mind when getting into arguments over monitor refresh rates. The average person cannot percieve motion at the rates some people claim.
The monitor refresh rate is associated to mouse curser aka cross hair.
And yes, we can perceive "motion of an enemy" much better if the refresh rate is higher.

Does a movie look better if the refresh rate is higher? Nope.

But if we play a game where my game engine is at 200 Hz and yours is at 60Hz: I kill you one shot every time I spot you.

Why is Palmer Luckey a "creep"?

Are you a TikTok girl or something?

Have you actually looked into his relationship habits?

He is a huge anime and retro video game nut. He married a cosplayer. (He did a little cosplay himself). They've been married since 2019 and have a kid together.

He won at life, basically.

We should really be building operating systems with more process isolation and clear permissions systems for which apps can interact with which other apps in which ways. There's elements of this in Android, and in capabilities systems like selinux, but the more powerful they get the harder they are to use. This is inevitable, but a lot more can still be done to make this kind of functionality accessible to users.

Not even Python could keep up with having to update from Python 2 to Python 3.