AI Will Transform the Enterprise. But Only Where the Foundation Already Exists.

The largest line item on most companies’ P&L is not infrastructure or software. It is people. Workforce spend accounts for 60 to 80 percent of total operating costs at most enterprises. It is the single biggest capital allocation decision a company makes, repeated thousands of times a year, across every department, every geography, every level.

Right now, every CHRO and CFO in the Fortune 2000 is being asked the same question by their board: what is your AI strategy?

The honest answer, at most companies, is that they are still figuring it out. Not because they are behind. Because the enterprise AI market is moving so fast that the best practices from six months ago are already obsolete. Buy a horizontal AI platform? Build custom agents? Wait for existing vendors to ship AI features? Deploy a coding agent and hope for the best?

I spend most of my time with the people who own or influence this cost center. And what I see is a gap the broader AI conversation ignores. Not between companies that have AI and companies that do not. Between companies that have the foundation for AI to work and companies trying to layer intelligence on top of a broken process.

The foundation problem

The AI industry has spent two years focused on model intelligence. Bigger models, better benchmarks, faster inference. The models have gotten very good. No argument there.

But model intelligence is only half the equation. The other half is what the model has access to. For most enterprise workflows, the answer is: not enough.

AI agents need three things to deliver real value in the enterprise. Structured, connected context. Native permissions and access controls. And a position inside the workflow where the work actually happens. Without all three, you get demos that impress and deployments that stall.

Most enterprise AI pilots plateau at proof-of-concept for exactly this reason. The model works. The context does not exist in a form the model can use. The permissions are not configured. The workflow integration is a sidecar that nobody adopts after the first month.

The bottleneck for enterprise AI is not intelligence. It is infrastructure. The model is ready. The enterprise is not.

Context is not a data problem. It is a systems problem.

When people talk about giving AI agents “the right context,” they usually mean data. Pipe it in from your systems, index it in a vector database, set up a RAG pipeline, let the AI retrieve what it needs.

That works for some use cases. It does not work for operational decisions.

Operational decisions require live, interconnected, structured data that reflects the current state of the business. Not a snapshot from last night’s data sync. Not an embedding of a document written six weeks ago. The actual state of things right now: which requests are pending, which budgets changed, which approvals were granted this morning, which plans shifted after yesterday’s reorg.

This kind of context does not live in a data warehouse. It lives in the operational system that governs the workflow. And for many critical enterprise workflows, that system either does not exist or is a collection of spreadsheets held together by manual processes and institutional memory.

Nobody talks about this. We talk about models and agents as if intelligence is the hard part. For most enterprises, the hard part is that the context the AI needs is scattered across five systems, two spreadsheet versions, a Slack channel, and the memory of someone who left the company last quarter.

Permissions will make or break AI adoption

In any enterprise above a few hundred employees, information access is not uniform. It is not supposed to be. A manager sees data about their team. A peer manager does not. Finance sees budget allocations. Recruiters see pipeline. An executive sees aggregate numbers. A front-line manager sees their department. Compliance, regulation, and organizational trust enforce these boundaries. That is by design.

Now introduce an AI agent that can query data across the organization. Who sees what? A manager asks the AI a question. Does the answer include data about a peer’s team? A recruiter asks about pipeline. Does the answer include the budget justification? An executive asks a broad question. Does the AI pull in compensation data that should be restricted to HR?

These are the first questions any CISO or General Counsel will ask when an AI deployment crosses their desk. The answer determines whether the project ships or sits in security review for two quarters.

Horizontal AI platforms handle this poorly. They either require a new permissions framework to be built, configured, and maintained from scratch. Or they punt on the problem and leave it to the customer. Both paths create months of delay and introduce risk that enterprise buyers will not accept.

The alternative is AI that operates inside a system where permissions are already enforced. Where security already approved the access control model. Where every query, every answer, every recommendation is automatically scoped to what that specific user is authorized to see. No additional configuration. No new governance project. No incremental security review.

For regulated industries, public companies, and any organization with more than a thousand employees, native permissions are the difference between an AI deployment that ships and one that dies in committee.

The sidecar problem

I see the same failure mode repeatedly. A company buys or builds an AI tool. It is powerful. It answers questions, generates analysis, surfaces insights. It lives in its own interface. Users go to it, provide context, interpret the output, then switch back to their actual system to do something about it.

Usage peaks in week two and declines from there.

The AI sits alongside the workflow instead of inside it. It creates additional work instead of reducing it. Users context-switch, copy-paste, and translate between the AI and their operational system. The friction is small on any single interaction. Compounded across a team and a quarter, it kills adoption.

The products that win in enterprise AI are the ones where intelligence is embedded in the system where the work happens. Not a separate tab. Not a chatbot floating in the corner of a different application. Native to the workflow, with the ability to surface answers and execute actions in the same environment where users already make decisions.

The companies that already own critical enterprise workflows have an asymmetric advantage here. They do not need to convince users to adopt a new tool. They need to make the existing tool smarter.

The best enterprise AI is the kind the user never has to go looking for. It is already there, inside the system, with the answer ready before the question is fully formed.

The governance layer is the control panel

Before AI can automate operational decisions in the enterprise, there has to be a governance layer that holds the context, enforces the permissions, and embeds the workflow. The AI is the intelligence. The governance layer is the infrastructure it runs on. Without the control panel, the intelligence has nowhere to go. It can think, but it cannot act. It can recommend, but it cannot execute within the constraints the enterprise requires.

This is true across every domain. Sales has the CRM as its control panel. Engineering has the codebase. Finance has the ERP and the general ledger. In each domain, AI is most powerful when it operates inside the system that already governs the workflow, holds the structured data, and enforces the rules.

For workforce decisions, this control panel has not existed. The ATS governs candidates, not headcount plans. The HRIS governs employee records, not workforce decisions. The FP&A tool governs financial models, not hiring approvals. Spreadsheets govern nothing. Each system holds a fragment. None hold the whole picture. None were designed to be the foundation for AI that automates the decisions upstream of all of them.

This is the gap we set out to close at TeamOhana.

We built the governance layer for workforce decisions: the system where headcount plans, requisitions, approvals, budgets, and org structures are structured, connected, and permissioned in one place. And we built Teemo, our AI Workforce Analyst, to operate inside that layer. Intelligence without infrastructure is just a demo.

The real question for CHROs and CFOs

The pressure to have an AI strategy is real. Boards want it. Peers are talking about it. Vendors are selling it. The temptation is to start with the AI: pick a model, pick a platform, run a pilot.

Start with the foundation instead. Before you evaluate any AI tool for any operational workflow, ask three questions.

First: is the context for this workflow structured and connected in a single system, or fragmented across multiple tools and manual processes? If fragmented, no AI fixes it. You need the governance layer first.

Second: are permissions already enforced in the system the AI will operate inside, or will you need to build a new access control framework? If the latter, add six months to your deployment timeline and a security review that may not end in your favor.

Third: will the AI operate inside the workflow where decisions happen, or will it live in a separate tool that requires your team to change how they work? If separate, expect adoption that peaks in the pilot and fades.

The companies that answer these honestly will spend less, deploy faster, and get more from AI than the ones chasing the newest model or the most impressive demo.

AI will transform the enterprise. But only where the foundation already exists. The governance layer is not a casualty of the AI revolution. It is a precondition for it.

The organizations that get this right will not be waiting for AI to catch up to their needs. They will have built the system AI needs to do its best work. Their people, from front-line managers to the CFO, will have something most organizations are still years away from: workforce decisions that are fast, accurate, governed, and intelligent by default.