Ross McLaughlin

Finding the constraint. Writing about AI, decision systems, markets, and the bottlenecks that move when technology changes.

[Ross McLaughlin @RealRossMc]

@garrytan Agree

[Ross McLaughlin @RealRossMc]

@joelmartinez @omooretweets What do you think automode is optimizing for? If routing is invisible, the platform is effectively deciding where deeper intelligence is worth paying for. Do you think users will eventually want control over that tradeoff?

[Ross McLaughlin @RealRossMc]

Naming agents around work beats recreating the org chart. Next comes mapping dependencies, especially where context or judgment crosses roles. I wrote about that here: https://t.co/xCOH48ST39

Does the watcher flag dependencies that should disappear, not just friction to automate?

[Ross McLaughlin @RealRossMc]

I’ve become pretty skeptical of org charts as a way to understand how a company actually works. The more useful thing is usually to follow the work and see what people really have to do to keep it moving.

Wrote about that here. https://t.co/CcJrOd8saw

[Ross McLaughlin @RealRossMc]

@antgrasso Exactly. The org chart shows where people sit, not what they do to keep work moving. I wrote about following the workflow instead: https://t.co/KRmQ4HBy7F

Are you seeing companies develop a reliable way to map that task redistribution?

[Ross McLaughlin @RealRossMc]

I’ve become pretty skeptical of org charts as a way to understand how a company actually works. The more useful thing is usually to follow the work and see what people really have to do to keep it moving.

Wrote about that here. https://t.co/CcJrOd8saw

[Ross McLaughlin @RealRossMc]

@themgmtconsult Companies map reporting lines, not the handoffs and judgment that get work over the line. I wrote about tracing the real workflow here:

Can it be mapped explicitly, or only seen when you try to change it?

[Ross McLaughlin @RealRossMc]

I’ve become pretty skeptical of org charts as a way to understand how a company actually works. The more useful thing is usually to follow the work and see what people really have to do to keep it moving.

Wrote about that here. https://t.co/CcJrOd8saw

[Ross McLaughlin @RealRossMc]

@illscience Once labs hide the model choice, the real product is the policy deciding where deeper reasoning goes, what gets checked and when the system stops. Will customers accept hidden logic on consequential work?

[Ross McLaughlin @RealRossMc]

I hit an AI usage limit this week and realized buying more capacity would only hide the real problem: the system was thinking too hard, too early.

I wrote about model routing, decision value, and where better judgment is actually worth paying for. https://t.co/fdIJW9gDDF

[Ross McLaughlin @RealRossMc]

@dair_ai This gets at the question I’ve been working on. Routing shouldn’t just pick a model. It should decide whether more reasoning is worth buying at all. I wrote about that here:

Could this extend across a portfolio of decisions, not just one query?

[Ross McLaughlin @RealRossMc]

I hit an AI usage limit this week and realized buying more capacity would only hide the real problem: the system was thinking too hard, too early.

I wrote about model routing, decision value, and where better judgment is actually worth paying for. https://t.co/fdIJW9gDDF

[Ross McLaughlin @RealRossMc]

@daniel_mac8 This is where things are heading. Th controller is difficult: what needs another model, when to escalate and when to stop. If Astra is trained end to end as the orchestrator, how do we audit those choices? I wrote about that layer here:

[Ross McLaughlin @RealRossMc]

I hit an AI usage limit this week and realized buying more capacity would only hide the real problem: the system was thinking too hard, too early.

I wrote about model routing, decision value, and where better judgment is actually worth paying for. https://t.co/fdIJW9gDDF

[Ross McLaughlin @RealRossMc]

@SenFettermanPA “AI supremacy” is doing a lot of work here. Does it mean the best models, more compute, wider adoption or the ability to set global rules? I wrote about why those lead to different strategies:

What outcome should the US actually optimize for?

[Ross McLaughlin @RealRossMc]

A lot of people are framing development of AI as a race to beat China, I was curious to understand what that actually means. https://t.co/oDlKcEXjUf

[Ross McLaughlin @RealRossMc]

@patrick_oshag If winning means decisive military superiority, China does not just accept second place. I tried to unpack the end states here:

Is a stable one-country lead even a realistic goal?

[Ross McLaughlin @RealRossMc]

A lot of people are framing development of AI as a race to beat China, I was curious to understand what that actually means. https://t.co/oDlKcEXjUf

[Ross McLaughlin @RealRossMc]

@alexwg This is why I don’t think AI is one race. The US can lead on capability while China wins on price and reach, and those advantages produce different outcomes. I tried to separate them here:

Which advantage do you think compounds more?

[Ross McLaughlin @RealRossMc]

A lot of people are framing development of AI as a race to beat China, I was curious to understand what that actually means. https://t.co/oDlKcEXjUf

[Ross McLaughlin @RealRossMc]

The general consensus is that leaders in AI have done a terrible job convincing people that it is a positive technology for society.

Yet I see very little being done by the same people to improve the PR of the industry. Why is that?

[Ross McLaughlin @RealRossMc]

A lot of people are framing development of AI as a race to beat China, I was curious to understand what that actually means. https://t.co/oDlKcEXjUf

[Ross McLaughlin @RealRossMc]

@therealpatkane @grok can you translate this into plain English?

[Ross McLaughlin @RealRossMc]

@justhefaxmam @garrytan Yeah, that’s worth factoring in. I’d just be careful with the 5M tons/GW number since it depends a lot on what’s actually generating the power.

Curious what generation mix you’re assuming there?

[Ross McLaughlin @RealRossMc]

@NY_LBSS @jaeporeon Loudoun specifically, not Virginia as a whole. Their numbers say data centers now make up 38% of general fund revenue and have helped keep property taxes lower. The statewide picture is a lot more mixed.

[Ross McLaughlin @RealRossMc]

@TheOculusOnline @jaeporeon I feel like efforts to establish expectations at this stage couldn't hurt people chances.

Uncommunicated expectations are just premeditated resentiments .

[Ross McLaughlin @RealRossMc]

@stelzner_n1150 Ship has sailed my guy

[Ross McLaughlin @RealRossMc]

@cremieuxrecueil Did some digging on the data out there for this

[Ross McLaughlin @RealRossMc]

I kept seeing the same arguments about data centers: too much water, too much power, too few jobs. So I went looking for the numbers. What I found was more mixed than the headlines. I wrote about what actually changes from project to project. https://t.co/88HzqFZYhI