us: we are struggling to figure out the best way to use coding agents, we don't have clarity yet
everyone else: our team is moving at speeds unheard of, all our PRs are ai generated, we've cleared 6 years of backlog
man we must really suck huh
Curious what American media is making of this, if anything.
the technology itself will endure. OpenAI doesn’t have to
Me: “Why are people so hostile to this interesting but unproven emerging technology?”
Me, after scrolling LinkedIn for ten minutes: “Oh.”
Yes. I've written hundreds of little one-off ETL scripts for various purposes over the years and now I generally just get an LLM to write it for me. Should I feel bad about generating fire-and-forget slop in a use case where output verification is easy?
“Too often dismissed as a consumer protection issue, data governance is a core condition of economic competitiveness and national security. Data sovereignty is ultimately about self-respect, and it’s way overdue.”
Good stuff from @vassb.bsky.social
www.theglobeandmail.com/business/com...
It also can't be from a US company. We're still mostly in the denial stage of the implications of the CLOUD act because the cost of action is so high, but either that shit gets repealed or we all have to stop doing business with, for example, Microsoft whether it's a domestic subsidiary or not.
He's right about privacy and security IMO. I don't work in a particularly critical sector but many clients have obligations around privacy and data sovereignty. If you're working with data that matters you can either buy inference by the token from a trusted non-US provider, or use your own GPU.
no actually im a tough guy clomping around in shoes three sizes too big you must respect me
Oh yeah Adelle. We were all disappointed, but were we really surprised?
It’s posts like this from Durbin that make it clear there is no real interest in repairing the relationship anywhere on the U.S. side. We won’t forget.
I'm sorry WHAT
enormously hot take: the reason a bunch of midwit software dev people are in their feels about agentic tooling is that it proves that good requirements specification and technical project management were the hard problem it's important for a human to be able to solve all along, not Coding™
Maybe there would be fewer white nationalists trying to get into the CAF and other places of influence in our society if we didn't look the other way while the world's richest man boosts white nationalism on the global platform he bought with the intention of boosting white nationalism.
Canada can start by making all politicians and government offices stop using X.
www.ctvnews.ca/politics/art...
We're currently trying to solve this for client work so I'm interested in your perspective. My current evaluation practice is rebuilding a working system from scratch using different approaches to the LLM. I'll try yours too.
I can see that. I think it's a real risk if you don't have a sense of what you want. I have experience leading developers who often have a similar tendency towards overbuilding, which I thin helps to mitigate this problem. I stress basic goals of minimalism, simplicity and human-readable code.
I haven’t been satisfied with the maintainability of LLM assisted code so I’m trying an approach of relatively long and detailed upfront conversations and maximum formality in terms of testing and linting. Way more than I require from human devs. Early days but I think the results are promising.
"Mainline conservatism is dying. In the US and UK at least, it is dead.This has been destabilizing to be sure, but let us not waste too much sympathy on it. It caused many harms and ignored many more. We should not fight for a creed that could not fight for itself."
LLMs are the latest in a long line of astonishing productivity gains for programmers that don’t change the essential nature of what we do, if that makes sense.
To your point there have been ways of churning out software at irresponsible speeds as long as I’ve been in the industry.
Software dev is a huge field. Many people look at how LLMs interact with their piece of the elephant, draw sweeping conclusions, then bring them to the public square as universal truth. Total dismissal and starry-eyed faith in the machine are both failure states.
I like how in 2026 a common security paradigm is writing a strongly worded letter to the guy in your computer
Dumbest thing Carney has done so far.
“We invoke international law and the “rules based international order” when adversaries engage in unlawful actions, but abandon those same rules entirely when it’s the Americans…For a country that depends on law more than force for its own security, that is not realism; it is recklessness.”
Oh I should have linked to the Mistral chatbot: chat.mistral.ai/chat
Mistral.ai is a European option worth considering. OLMO from the Allen Institute (playground.allenai.org) is an LLM trained on ethically sourced public data.
Valid perspective. Big tech, especially US tech, is a risk.
Not at all. Just an off the shelf LLM, not even one of the big frontier ones. I use open models, either on my own hardware or a European inference provider.
This sort of work is pretty tedious to do by hand, but tweaking output that is 80-90% there is a lot faster and more fun.
Sure. I was working on a bid for a public contract recently.
I loaded the bid doc into visual studio and chatted with an LLM using Cline in plan mode until I was satisfied it understand the goal. I got the LLM to generate planning artifacts for me: schedules, labour worksheets, stuff like that.
