Delaware's AIC gets it backwards. They're proposing to wrap an agent in a legal entity that is accountable to even law suits. If you look at the history and purpose of LLCs, you see a structure that encourages entrepreneurship by shifting accountability from someone wallet to their conduct.
Fiduciary duty, negligence, and veil piercing all need a persistent decision-maker. An agent-operator has no mens rea, no continuity between incarnations, no causal model to be deterred through.
As proposed, the AIC severs liability from judgment. And since agents are somewhat interchangeable, putting a different skin on one and creating a new AIC after a failed one likely won't adapt conduct.
Delaware is considering building the wrong primitive first.
8
"Being AI-native" is a posture, not a strategy. Strategy is the list of things you're refusing to build.
2
40
How many frontier models are there actually if Meta, SpaceX, Thinking Machines, and Moonshot keep releasing competitive models at much cheaper prices?
15
The framework asks juniors to form a point of view before opening the AI tool. A first-year doesn't have one yet.
The pilot's star case was a consultant who "had never worked without AI." That's the person who most needs judgment having the least of it.
"What looks right but isn't" is the costliest AI error, and it needs context a novice hasn't earned. You can't diagnose a stale-data trap you've never been burned by.
Old apprenticeship was slow because you wrote the bad draft and watched the client flinch. Remove the drafting and you remove the thing that built the judgment the AI now demands as input.
A reasoning trail audits judgment. It doesn't manufacture it. Great for the partner, silent on the analyst.
1
52
The biggest blocker to realizing AGI is unit economics of experience (aka the feedback loop models need to really learn).
13
People who have a high amount of agency in their professional lives feel invigorated by AI while people who are stuck in legacy organizational constructs feel burnout
1
1
56
23% of companies already list AI agents on the org chart. According to a BU study, the title is a branding exercise that makes the humans around it measurably worse at their jobs.
2
64
Effective Immediately, all organizations using PHP for their systems are advised to be prepared for an update addressing two critical vulnerabilities. PHP is widely used for web applications. These vulnerabilities often provide a direct path to the server's file system or database. If an immediate update is not feasible, it is recommend that organizations take the system offline until it can be fully mitigated. NSA will follow up shortly with more specific guidance.
Meta's CTO, Andrew Bosworth says owning your own model backstops how much frontier labs can charge you. He also says most human tasks run fine on conventional intelligence. Maybe pick one? If the second is true, DeepSeek and Qwen are already a free backstop.
1
38
The labs' escape from the commodity trap runs straight into a worse one: vertical value capture pays services margins, not software margins.
16
The SaaSpocalypse framing is too blunt. Sanofi cut 80% of ServiceNow with Claude agents and went deeper on SAP in the same motion.
Agents make workflow the cheapest thing to regenerate. Verified state is the scarce asset.
40
Seamless AI adoption measures adoption. It doesn't measure whether the people using it are getting better or worse at their jobs.
UChicago Law is banning devices from 1L classrooms and adding mandatory oral defenses for research papers — deliberately, to re-engineer friction back into the learning process.
AI shortcuts are a feature in professional contexts and a defect in learning contexts. Most enterprise deployments never ask which one they're in.
If judgment is where differentiation lives — in law, in strategy, in any high-stakes domain — the ROI question for AI isn't just time saved. It's whether the humans in the loop are sharper or duller six months later.
1
38
"Every business will have to be re-founded as a software business" as AI codifies tacit operational knowledge. I can see the value of that, but Tacit knowledge exists precisely because it resists articulation
notboring.co/p/return-on-tok…
11
The functional org chart — eng, PM, design, DS — was always an approximation of how product work actually gets done.
I've watched this dynamic play out inside large engineering orgs. The best teams self-organized around what the product needed, not around functional headcount.
nitter.cf/bcherny/status/2071379…
As engineering, product, design, DS, etc. melt into a new kind of role, I was reflecting on what roles might look like in the future. For example, when I look at the Claude Code team I see what I think is five archetypes:
1. Prototyper: comes up with brand new ideas; churns out many ideas, most of which don't ship
2. Builder: quickly turns a prototype/idea into production-grade product/infra
3. Sweeper: cleans up the UI, simplifies the code and system, unships, optimizes performance
4. Grower: takes a product that has been built and iterates on it to improve Product-Market Fit
5. Maintainer: owns a mature system to make it secure, reliable, fast, and efficient as it scales
Many people span across 2 roles, and sometimes 3 roles. I also notice that these roles are not really tied to job function -- eg. across Anthropic, some designers match category 1, some 2, some 3; same for engineers, PM, DS.
A healthy team needs a mix of these, depending on the product:
- A product that is new and pre-PMF needs people that are strong at 1+2+3
- A product that is growing and has found PMF needs 2+3+4 and some 5
- A product that has strong PMF needs 3+4+5 and some 2
Maybe product roles of the future will look more like this, and less like the domain-specific roles of today?
9
Meta reportedly building a cloud compute business to sell excess AI infrastructure to outside developers shows a significant hedge buried in a very bullish CAPEX narrative.
bloomberg.com/news/articles/…
17
Model routers are having a moment but don't forget that router choice is an architectural opinion, not a neutral infrastructure or procurement decision.
Some routers optimize for cost. Others for performance. Some let you tune a dial between the two but you have to design that tradeoff upfront, before you know your actual production task distribution.
Treat router selection like any other consequential design decision: know what you're optimizing for, who owns that tradeoff, and what it costs to revisit it.
theinformation.com/newslette…
1
25
Alex Karp tapped into the frustration around token economics and data privacy.
Enterprise AI budgets are blowing past projections. Foundation model providers keep launching products that edge closer to their customers' workflows. And what the contracts say and what the CTO believes is happening feels like it's widening.
If American open-source models reach capability parity for classified defense use cases, proprietary model providers lose a significant wedge of what they may have considered strategically protected revenue.
cnbc.com/2026/07/01/palantir…
55
"Skill engineering" is a more useful frame than it might sound.
Prompt engineering tells an agent what to do. Skill engineering gives it the domain knowledge to do it well — vocabulary, context, professional judgment compressed into the system.
The gap between a designer and an engineer using the same model usually involves the vocabulary each brings to it.
36
Coding agents increased developer commits by 180% but releases only by 30%.
We're accelerating the wrong part of the workflow.
exponentialview.co/p/the-ai-…
23
LLMs got lucky with a ready-made training set. The internet was already full of text.
World models have no such gift. Geometry, physics, and material properties aren't just sitting there waiting to be scraped.
Most 3D data is locked in proprietary CAD files. Real physics demonstrations are expensive to generate at scale.
worldlabs.ai/blog/post/2025-…
1
33



