AI can increasingly generate alternatives, simulate outcomes, and execute workflows, but determining objectives, acceptable constraints, decision authority, and what constitutes a good outcome remains essential.
The hardest problem in enterprise AI is increasingly not reasoning about a task. It is defining the right task from the full state of the business—and for many of the most valuable decisions, that state extends beyond the enterprise boundary.
What we have been building at SAP Business Network gives AI the network state, authority and execution environment required to optimize the right decision rather than merely automate the task it was given.
Let us walk through an example:
A technically excellent agent might be given the task:
“Expedite the late purchase order.”
It could execute that task perfectly and still make the business worse.
The actual task might be:
“Protect the customer promise at minimum expected economic or esg cost, given DEEP NETWORK supplier capacity, supplier component availability, alternate sources, logistics options, contractual commitments, and within company inventory policy and downstream production constraints.”
Those are radically different optimization problems and AI ROI impact.
And critically, the second task cannot be understood from the enterprise system alone. It requires network state.
Choosing the right task often requires leaving the AI problem entirely to study users, products, organizations, or whatever part of the enterprise is supposed to benefit.
Once the task and data are right, scale is incredible. Then you are on your way to scaling based on Sutton’s Bitter lesson.
Hello, world. We've been building a terminal multiplexer and we think you're going to love it. Native apps (e.g., macOS/iOS), accessible from the web, multiplayer. From the creators of Ghostty and Echo.
https://t.co/VRdIMWMLHL
@PalantirTech In basic terms they just wrote code instead of use AI. Code is still 10000x faster if it can be used then having AI compute to do the same things. Not sure they are aware of it or are hyping some kind of financial scam; perhaps just low IQ if unaware. Yes, code is faster than AI.
Kimi K3 Isn’t the Story. Margin Compression at the Foundation Model Layer Is. This actually strengthens the software layer
Ironically, cheaper foundation models make enterprise software more valuable.
The bottleneck shifts.
Winning AI products won’t simply wrap an LLM.
They will own:
* enterprise context
* proprietary workflows
* trusted data
* evaluation pipelines
* governance
* decision logic
* execution systems
* network effects
This is exactly why I’m increasingly convinced the competitive advantage for enterprise software isn’t the model.
It’s the decision layer that surrounds the model.
One thing that’s become clearer to me over the last couple of years building enterprise AI systems…
We’ve spent a lot of time asking, “Where should we use an LLM?”
I’m starting to think the more important question is, “Where shouldn’t we?”
The pattern we’ve gravitated toward is using LLMs to understand people—intent, ambiguity, conversation, recommendations.
But once you cross into executing business logic, making decisions, querying enterprise data, applying policy, or taking action, we try to get back onto deterministic ground as quickly as possible.
Semantic models.
Knowledge graphs.
Business rules.
Optimization.
Workflow.
Governance.
Those aren’t limitations. They’re what make enterprise software trustworthy.
I think a lot of AI products today still have an LLM sitting in the middle of everything. It demos well, but it also introduces variability, cost, and makes governance harder than it needs to be.
My guess is the next generation of enterprise AI architectures will look different:
Use probabilistic AI to understand what people mean. Use deterministic systems to decide and execute.
It also changes the economics.
If every interaction doesn’t require another expensive reasoning step, you stop thinking about tokens and start thinking about business outcomes.
Curious if others building enterprise AI are seeing the same thing, or if you’ve landed somewhere different.
Hosted a dinner last night with a group of IT leaders of large enterprises around agent adoption in the enterprise. Some quick notes:
* Change management remains one of the biggest topics for driving workflow transformation. Still most processes need to be upgraded to modern operating models to work with agents, which is a mix of technology, data, and human process change. Lots of emphasis on getting data (structured and unstructured) into a setup that can work with agents properly.
* IT teams are finding increasing success embedding full engineers into the business functions (essentially internal FDE) that go and implement agents into the internal workflows. There’s so much technical work to be done to make agents successful, that they can accelerate months or quarters of failed experiments by having someone technical in the workflow early.
* Consensus that the tech function is becoming more important than ever. It’s clear that the business could only expect automation to affect a minority of the business before (e.g. ERP) but now it can impact all of knowledge work. This means IT is becoming a more central role to the workflows across the company.
* Workflows are cross functional, and getting agents to work cross functionally is a complicated data modeling and permissions issue. Single users don’t have access to this. Which means you need to have agentic systems take on their own roles and have their own privileges, which is non-trivial given agents can’t keep things secure on their own.
* Huge variance in budgets between coding work and the rest of knowledge work. Some companies had a $1,000 a month budget for developers, and others had much higher amounts (like $5,000) that were merely triggers to notify the team vs. block them. Far smaller budgets for non-coding work at the moment.
* More companies are building their own multimodel systems for routing workloads by task to frontier and lower cost models. Lots of energy around open weights models, but still more in experimentation instead of at scale usage (some companies can’t due to perceived Chinese issue).
* Clear sense that all enterprise software must be headless in the future. Relief that they don’t need to train employees on hundreds of different apps. However, clear frustration with the traditional vendors that don’t play extremely nice (technically or cost wise) with agents in a headless fashion. Huge warning for existing software vendors.
* Mythos or mythos level-models are finding more and more sophsiticated security risks. The chaining together of vulnerabilities is what’s novel right now, and companies are coming up with long backlogs of what they need to go patch quickly.
Even more discussed, but just a few of the hottest topics.
Agentic AI is exploding in engineering — but real winners will deploy it for specific workflows/business processes with deep domain expertise.
Pair builders with experts. Redesign around humans + AI. Balance with decision intelligence: formal modeling & science-based decisions.
Success demands FDE support, change mgmt, clean data, robust evals & more. Keep humanity centered. 🚀
As frontier models keep improving and (to some degree) commoditize, the winners in agentic AI will likely be the companies that best solve the context + integration + governance + change management problem for real work. Pure model labs have the intelligence; the applied layer has the environment where that intelligence actually gets useful work done.
250 years ago, a group of imperfect but courageous people made a radical bet: that a nation could be built not on kings, bloodlines, or inherited power, but on an idea.
The idea that liberty matters.
That people are born with rights no government can give and no government should take away.
That self-government is hard, messy, and unfinished — but worth fighting for.
America has never been perfect. But that is part of what makes this country extraordinary. We are a nation constantly arguing, correcting, building, rebuilding, expanding the promise, and reaching again for the ideals written into our founding.
We helped defeat tyranny in world wars. We rebuilt nations after war. We protected allies. We opened doors for immigrants, entrepreneurs, dreamers, builders, scientists, artists, and ordinary families chasing a better life.
We put humans on the moon. We built the world’s most dynamic economy. We created technologies, medicines, companies, universities, music, movies, and ideas that shaped the modern world.
But the greatest American achievement is not any single victory, invention, or institution.
It is the belief that tomorrow can be better than today — and that free people, with courage and responsibility, can make it so.
On this 250th Fourth of July, I feel deep gratitude for the country we inherited, pride in what America has done for the world, and responsibility for the country we pass on next.
Happy Independence Day.
May we remain worthy of the idea.
This is what’s causing Anthropic to aggressively beg for govt protection (see below). Customers are finding cheaper alternatives. Keeping employees requires continuing ultra-rich secondaries ($$$) that are dependent on revenue growth. When you can’t win on the field go to DC.
What once began as a noble experiment in forming a digital currency unencumbered by financial institutions, cryptocurrencies have proven to be an abject failure by every measure.
There are on the order of only 30 million bitcoin wallets whose value is greater than $100 US; this represents at best about 0.36% of the world’s population, hardly a sufficient number to provide a global medium of exchange. While there certainly exist some fringe cases of utility, most notably by individuals needing to transport value outside dictatorial countries with broken and oppressive economies, cryptocurrencies are largely an object of speculation, driven by market manipulation and the economy of the greater fool and favored by criminal organizations and countries such as Russia and North Korea seeking to avoid sanctions. The ancillary industries spawned by cryptocurrencies – NFT and DeFi and so-called Web3 - never gained traction, and indeed proved to enrich a small few at the expense of the many.
At the start of this technology, I was curious and hopeful, but it is clear that human greed has irredeemably ruined it for all.
He spent 25 years inside one company.
No personal brand machine.
No job hopping flex.
No podcast tour career strategy.
Just decades of compounding competence.
“Quiet people often look average right before they look inevitable.”