Tassilo Klein

[email protected]; Principal Research Scientist; ex-SAP, ex-postdoc @Harvard@MIT; NLP, computer vision; views are my own.

[Tassilo Klein @T_J_Klein]

Corporate hierarchy is a 2,000-year-old information routing protocol.
AI just made it optional.
Two independent groups figured this out separately, one from economic theory, one from running a company at scale.
https://t.co/mcUoPIt6St
@roelofbotha @sequoia
#FutureOfWork #AI

[Tassilo Klein @T_J_Klein]

In 2009, every e-commerce company was rebuilding payments from scratch.
Then Stripe happened.
The same moment is arriving for enterprise AI. The layer nobody is building yet is the one everything else depends on.

https://t.co/QR6pPLluc0

#AgenticAI #EnterpriseAI

[Tassilo Klein @T_J_Klein]

3/ Full piece here: https://t.co/bYpsr5mtyO

[Tassilo Klein @T_J_Klein]

AI agents are being deployed into enterprise environments they cannot see.
Here is why that is the core problem nobody is talking about, and why it matters more than which model you use: 🧵

[Tassilo Klein @T_J_Klein]

2/ 71% of enterprise applications remain unintegrated. 82% of business leaders say AI fails without understanding how the business runs. The gap is not a model problem. It is a process visibility problem.

[Tassilo Klein @T_J_Klein]

@garrytan MCP describes tools. It tells agents nothing about how to use them safely in context. Agents stay blind to execution reality until the protocol closes that gap. 100LOC wrappers win because they already have.

[Tassilo Klein @T_J_Klein]

The mediocrity tax is dead.

For 40 years, enterprises paid premiums for integrated suites because custom integrations were too expensive.

LLMs changed the math. Coordination is now cheaper than containment.

What comes next: https://t.co/PgVrI5JesX

[Tassilo Klein @T_J_Klein]

@saen_dev Not to forget all the rules and role management coded into the UI…

[Tassilo Klein @T_J_Klein]

Full paper: https://t.co/1JRT3E0fjY

[Tassilo Klein @T_J_Klein]

For 40 years, enterprise software was governed by a single economic law: integration beats specialisation. That law just broke. A thread on why, and what replaces it.

[Tassilo Klein @T_J_Klein]

The model is architecturally clear. It is not yet operationally simple. Most agents being deployed today cannot actually see how enterprise software works in production. That gap is the most important near-term problem in enterprise AI, and it is underappreciated.

[Tassilo Klein @T_J_Klein]

Excited to be at #NeurIPS2025 in San Diego!
If you’re into foundation models for tabular/relational data, multi-agent simulation, or synthetic data generation, let’s connect.
Also: SAP is hosting a reception — join us!
👉 https://t.co/D79CiSII8u
#AgenticAI #EnterpriseAI

[Kayhan Batmanghelich @kbatmang]

We’re excited to announce that our MedSyn project is finally out after 2 years of dedicated work. Congrats to @yanwuxu6 and Li Sun!

MedSyn is a fully controllable Generative AI that synthesizes anatomically realistic CT images using text prompts with extensive evaluation.

[Julian Togelius @togelius]

Six years ago, @yannakakis and I published a textbook on AI and Games. It has been used in universities across the world and become a standard reference for the field. Much has happened since, and it is about time for a second edition. Good news: we just put a draft online!