Sam Altman has acknowledged that AI adoption has been slower than expected, while Meta has reportedly struggled with efforts to reorganize parts of its workforce around AI agents. The two developments point toward the same problem: the technology is advancing faster than organizations are learning how to deploy it.
The challenge is no longer simply building an AI system that can perform a task. It is integrating that system into an organization where processes are fragmented, exceptions are everywhere and much of the knowledge required to make decisions never exists in a database.
Sam Altman has acknowledged that AI adoption has been slower than expected, while Meta has reportedly struggled with efforts to reorganize parts of its workforce around AI agents. The two developments point toward the same problem: the technology is advancing faster than organizations are learning how to deploy it.
The challenge is no longer simply building an AI system that can perform a task. It is integrating that system into an organization where processes are fragmented, exceptions are everywhere and much of the knowledge required to make decisions never exists in a database.
Sam Altman has acknowledged that AI adoption has been slower than expected, while Meta has reportedly struggled with efforts to reorganize parts of its workforce around AI agents. The two developments point toward the same problem: the technology is advancing faster than organizations are learning how to deploy it.
The challenge is no longer simply building an AI system that can perform a task. It is integrating that system into an organization where processes are fragmented, exceptions are everywhere and much of the knowledge required to make decisions never exists in a database.
The Gap Between AI and the Enterprise
The Gap Between AI and the Enterprise
The Gap Between AI and the Enterprise
AI models can now generate code, analyze information and perform increasingly complex tasks. But businesses rarely operate through clean, predictable workflows.
Information is distributed across systems. Processes contain exceptions. Decisions depend on institutional knowledge. Much of what makes an organization function is never formally documented.
That makes deployment considerably harder than the product demonstration suggests.
Winners in AI will not be the ones with the flashiest demos, they will be the ones who turn messy work into reliable workflows.
Winners in AI will not be the ones with the flashiest demos, they will be the ones who turn messy work into reliable workflows.
Winners in AI will not be the ones with the flashiest demos, they will be the ones who turn messy work into reliable workflows.
This is where the forward-deployed engineer, or FDE, becomes increasingly important.
Rather than building AI products from a distance and leaving customers to integrate them, FDEs work directly within organizations. They understand existing workflows, identify where systems break down and adapt AI to the realities of the business.
Diagram of forward-deployed AI workflow.
Diagram of forward-deployed AI workflow.
Forbes reports that FDE hiring has expanded significantly across the technology industry, including at companies such as OpenAI, Anthropic, Google and Salesforce.
The role reflects a broader shift in AI: as models become more capable and accessible, the competitive advantage may move away from the model itself and toward the layer that connects it to real operations.
Forbes reports that FDE hiring has expanded significantly across the technology industry, including at companies such as OpenAI, Anthropic, Google and Salesforce.
The role reflects a broader shift in AI: as models become more capable and accessible, the competitive advantage may move away from the model itself and toward the layer that connects it to real operations.
Forbes reports that FDE hiring has expanded significantly across the technology industry, including at companies such as OpenAI, Anthropic, Google and Salesforce.
The role reflects a broader shift in AI: as models become more capable and accessible, the competitive advantage may move away from the model itself and toward the layer that connects it to real operations.
From Model to Operating System
Real estate makes this particularly clear.
A modern property generates vast amounts of information: equipment readings, maintenance requests, energy consumption, access events, contracts and tenant communications.
An AI system can identify an abnormal HVAC reading, for example. But identifying the anomaly is only the beginning.
Is the equipment under warranty? Has the issue happened before? Is the building occupied? Which contractor should respond? How urgent is the intervention?
The intelligence becomes valuable when it can connect the signal to the decision.
That is the opportunity for AI in real estate: moving from systems that report what is happening to systems that understand what should happen next.
The AI industry has spent years competing to build increasingly intelligent models. The next competition may happen somewhere less visible: inside the workflows where those models are deployed.
Sam Altman and Meta's experiences suggest that the difficult part of enterprise AI may not be capability alone. It is integration, context and trust.
For businesses, the question is shifting from “What can AI do?” to a more consequential one:
“What would we need to change for AI to work here?”
That is where the next phase of enterprise AI will be decided.
