@ashwingop Amdahl's law is the perfect frame for this. Making one person in a ten-person chain 100x faster so the task finishes 2% sooner should end every 'AI made my team faster' story. The company is the unit of analysis, not the employee.

The memory is the hard part to get right, especially at enterprise grade with access control, governance etc. Which we already have figured out and in production. Its hard to explain how magical agents can get with the right memory architecture underneath it.

Now, imagine this in an enterprise context? Imagine an agent that has continously updating understanding of what's happening everywhere in your org, bringing you insights you didn't even know existed. Using personal agents for content consumption and personal usecases are great. Its impact on enterprise will be orders of magnitude more.

A $12B public company (@Lenskart_com) got measurably faster across ten depts without a single employee getting faster. This is the kind of AI results that impacts P&L, without the theatrics of tokenmaxxed leaderboards and here's the arithmetic behind it.

LMAO.. Dario Amodei wants governments to call Beijing and put a “speed limit” on AI. It sounds more like: “Please ban or slow them down because they’re hurting our business.” People are running DeepSeek v4.1 Flash on two DGX Sparks and it’s mogging our best closed models.

EA AI safetyism increasingly looks like Marxism-Leninism for the algorithmic age. The old vanguard claimed privileged knowledge of the inevitable course of History. The new one claims privileged knowledge of the probabilistic course of Humanity. Both use an elaborate intellectual framework to reach the same political conclusion: a small group of enlightened people must constrain everyone else for their own good. That has never lead to anything except monumental human suffering.

Every company has the same problem says @ashwingop. One team learns something. Nobody else finds out. The lesson gets relearned six months later by someone else. "How do you make sure that if one part of the company learns about something, everybody else learns it too? So you don't relearn experiences. You move faster. You move as a single unit." That's what @sentra_app is building: organization memory. A company brain. Full convo below.

How do you introduce AI to a child who has already interacted with it but has no idea it's not a person? @Ashwingop's daughter started talking at @ChatGPT the same way she talks to her grandparents on the phone. She couldn't tell the difference. Full convo below

Instinct and Muse can do our work, they just can't do it in a real company. We plugged both into Sentra (@sentra_app), our collective memory platform. They could do the work, but not the way the company would. They were missing its context, operating rules, and institutional memory. So we built the "hard part" ourselves and gave our memory agent a browser. As it turns out, LLM does most of the heavy lifting in browsing. The actual hard part was the memory, and that part is already running in several mid-to-large companies, including one with 20,000 employees. The governance mostly follows from it: something that knows what the company knows also knows what it isn't supposed to do. A first glimpse of what's coming: collective org memory that can act, and stays controlled. It's like watching a child take its first steps. Tiny, and probably magical only to the parent. But even Usain Bolt started with one seemingly insignificant step :-)

BYOD (Bring Your Own Devices) to work is a nuisance to IT. But, BYOA (Bring Your Own Agent) to work will break the enterprise. Instinct and Muse are incredible for individual tasks, making it tempting to assume they will just work across an organization. They will at first in small numbers, but as personal agents proliferate, operations will rapidly degrade into chaos without a unified shared reality for coordination.  Personal agents thrive because they have rich personal memory and simple situational awareness. However, in an enterprise they will have only limited awareness without much needed organizational memory: a unified memory/state layer with fact-level ACLs and governance so agents can actually work together without tearing the company apart. Our deep dive on the emerging BYOA problem

“My agent did it”, I bet this excuse is going to be used a lot more an instinct and muse agents proliferate. The memes and general lunacy waiting for us in the coming days is going to be glorious :-). “my agent didn’t do it” might also work equally well.