It’s about the agent. Not the model or the format.
The "bitter lesson" only applies if the model makes the agent redundant. We aren't there yet. Agentic loops are just software engineering on top of CS constructs; they help current models produce better results.
Could models eventually internalize the logic used in Claude Code / Codex / OpenCode / Aider? Maybe. But for now, keeping that complexity in the agent is more energy-efficient. Even if complex agents eventually get replaced by simple loops, these standards save tokens and time today. That’s worth something.
that's true, but hardly relevant for the grid, especially considering many ppl dont even have a battery and have grid following solar panels which shut down the moment grid is down
Hi Peter, let's play hypotheticals. USMCA (CUSMA) is dead, Trump actually acted on his threat and the US left the treaty.
I assume it means TN visas are dead in six months if the written notice clause is respected.
Is the zombie CUSFTA now out of dormancy? Can we now use the TC visa?
If CUSFTA dies as well, what are the other options and how long does it take to process?
Just my opinion, but I think USMCA dying is massively unlikely. Already there is some resentment in the Republican base for souring ties with Canada (e.g. bourbon producers, tourism) and I do see Canada-US ties moving in an upward trajectory in the recent months. The USMCA is also used by US citizens to work in Canada. And considering how watered down the H1B changes have been, I don’t think this administration has an appetite for materially pissing off coastal hi-tech. At most we’ll see tightened category adjudications.
I persuaded my bank out of $200 using AI to formulate the formal ask using their pdf as guidance. I could have gotten it directly but the effort barrier was too high for it to be worth it.
However, as soon as they put AI to handle these queries, this will result in having AI persuade AI. Sound like we need a new LLM benchmark: AI-persuasion^tm.
"This fund, which we like to refer to by its trade ticker, “[NANC/GOP],” invests in equity securities purchased or sold by [Democratic/Republican] members of Congress and their spouses
Not sure but there used to be a Nancy Pelosi tracker account on Twitter, and they would follow those trades, person made a bit of money, nowhere near what Nancy's estate has made though.
At some point our cross-AZ traffic for Elasticsearch replication at AWS was more expensive than what we'd pay to host the whole cluster replicated across multiple baremetal Hetzner servers.
Could we have done better with more sensible configs? Was it silly to cluster ES cross-AZ? Maybe. Point is that if you don't police every single detail of your platform at AWS/GCP and the like, their made-up charges will bleed your startup and grease their stock price.
turns out cross AZ is recommended for ES. perhaps our data team was rewritting the indices too often. but it was an internal requirement. so I think the data schema could have been more efficient to append deltas instead of reindexing all. but none of that will inflate your bill significantly at Hetzner. of course it will at AWS as that's how they incentivise clients to optimize and reduce their impact. and that's how you cut your runway by 3-6 months in compute heavy startups
Managed servers reduce the on-prem skillset requirement and can also deliver a lot of value.
The most frustrating part of hyperscalers is that it's so easy to make mistakes. Active tracking of you bill is a must, but the data is 24-48h late in some cases. So a single engineer can cause 5-figure regrettable spend very quickly.
I've seen this in startups with 7 figure ARR (where annual cloud costs were also 7 figures).
Also seen that in F500 where a single architect caused a 5-figure mistake which remove cloud privileges from the entire architecture team. Can't make it up.
7 figures is single digit millions. 5 figures is a single silly mistake. Just enabling more verbosity on logs is enough to trigger that in a day. Such mistakes would be found monthly. We could have easily had 10% more engineers with the same budget if it weren't for lighting our runway on fire that way.
Ultimately these owners hire me to cut their 6-figure AWS bill by 50%. It's mostly rearchitecting mistakes. Amongst them is taking AWS blog propaganda at face value. Those savings could be 80% if they chose managed bare metal (no racking and stacking).
The "bitter lesson" only applies if the model makes the agent redundant. We aren't there yet. Agentic loops are just software engineering on top of CS constructs; they help current models produce better results.
Could models eventually internalize the logic used in Claude Code / Codex / OpenCode / Aider? Maybe. But for now, keeping that complexity in the agent is more energy-efficient. Even if complex agents eventually get replaced by simple loops, these standards save tokens and time today. That’s worth something.
reply