Your Company Has AI Everywhere. So Why Isn’t It Getting Faster?

Written by: Bill Mastin

Walk through any mid-size company today and you'll find AI in almost every department.

Recruiting has AI screening résumés and scheduling interviews. Sales has AI writing outreach, scoring leads and summarizing calls. Support has AI deflecting tickets and drafting replies. Finance has AI matching invoices, flagging anomalies and speeding up the close. Marketing has AI generating copy, variants and campaign reports. Engineering has AI writing code, reviewing pull requests and triaging incidents. Legal has AI redlining contracts. Operations has AI forecasting demand and routing work. HR has AI answering benefits questions and drafting reviews.

Each of these tools shows a real gain on its own. The task gets faster, and the demo is impressive. The vendor can point to hours saved.

Yet many leadership teams are stuck on the same uncomfortable question: if every function is faster, why doesn't the company feel faster?

MIT's Project NANDA put a number on that feeling in its 2025 report, The GenAI Divide. Across the enterprise pilots it studied, roughly 95% showed no measurable P&L impact. The researchers didn't blame model quality. They pointed to a "learning gap": most tools don't retain feedback, don't adapt to context and don't improve over time. In other words, they have no memory of the business they're working in.

My argument is that this is a structural problem. We have been giving the company a hundred separate brains when what it lacks is a nervous system.

The point solution trap

Point solutions optimize tasks. Companies produce value through flows.

A deal is not closed by the sales AI. It moves from marketing to sales to legal to finance to onboarding to support. At every handoff, context leaks. What the customer said on call three, why the discount was approved, which promise the account executive made, what the implementation team already knows: these live in someone's head, a Slack thread or a meeting nobody recorded.

Speeding up one step in that chain does very little if the constraint sits somewhere else. That is the core lesson of Eliyahu Goldratt's Theory of Constraints: an hour saved at a non-bottleneck is a mirage. In most companies, the bottleneck is not how fast individuals do their tasks. It is how fast shared understanding moves between people, teams and systems.

So we get a predictable pattern:

Local speed, global drag. The sales AI writes 5x more emails, and the pipeline review is still a two-hour meeting spent reconstructing what happened.
Context re-creation tax. Each AI tool starts from zero. People spend their saved time re-explaining the situation to the next tool, or the next colleague.
Divergent truths. Five AI tools trained on five slices of data produce five versions of the customer, the product and the plan.
Amplified misalignment. When a team is out of sync, AI helps it go faster in the wrong direction.

AI point solutions make individual functions smarter. They don't make the organization more coherent. Coherence is where the compounding returns are.

Sync: the missing performance variable

This is where the idea of organizational sync comes in.

Sync is the degree to which the right people and systems hold the same current, accurate context at the moment they need it. It is not alignment in the strategy offsite sense. It is operational: does the person or agent about to act know what the rest of the company already knows?

The conclusions so far:

The legacy baseline is a siloed enterprise. Context is stored in function-specific systems and passed along by meetings, email and memory. Sync is low, and keeping it up costs a lot of human effort. Much of middle management exists to be the human routing layer for context.
Sync can be measured. Rather than a purely qualitative story or a single hard metric, the approach is a scored index on a maturity ladder. It scores factors like context latency (how long it takes a fact to reach everyone who needs it), handoff loss, decision rework and time spent searching or re-explaining.
The ladder describes an evolution: Siloed: context lives in functions and heads, and sync depends on meetings. Connected: systems are integrated, but people still have to go find the context. Augmented: AI point solutions speed up tasks within functions. Most companies are here now. Contextual: a shared context layer feeds every tool and person the same current picture. Autonomic: the organization maintains its own sync in the background, the way a body regulates itself.
The hypothesis to test: moving up the ladder improves efficiency more than adding point solutions within a level. Most of today's AI spending sits on the third rung, and the next jump comes from the fourth and fifth rungs.
Theory of Constraints makes it a continuous improvement loop. Measure sync, find the flow where context loss is the constraint, raise sync there, then find the next constraint.

The claim is simple: sync is a performance multiplier, and AI without a context layer caps out at the Augmented level.

The company brain is a context layer

If sync is the goal, the company brain is how you get there.

I don't mean a chatbot sitting on your wiki. I mean a context layer that sits under every function and every AI tool. It holds a living model of the organization: customers, commitments, decisions, projects, people, expertise, and the reasoning behind all of them.

Every point solution plugs into it. The sales AI drafts a proposal knowing what support heard last week. The recruiting AI screens candidates knowing what the hiring manager actually valued in the last three hires. The finance AI flags an unusual contract term knowing that legal already approved it and why.

The point solutions stay. They finally share one understanding of the business.

Why it has to be autonomous and subconscious

Here is the part of the thesis I think matters most.

Knowledge management has been tried before. Intranets, wikis, SharePoint, "document everything" mandates: nearly all of it slowly decayed. The reason is always the same. It depended on humans doing extra work to feed it. Documentation is a tax on the people who are busiest, so it is always stale, and people stop trusting stale knowledge.

If the company brain works the same way, it will fail the same way.

The human body offers a better model. You don't consciously manage your heart rate, blood pressure, digestion or balance. The autonomic nervous system handles these in the background, so your conscious mind can focus on judgment, creativity and relationships. Reflexes move signals to where they are needed before you have time to think.

An enterprise nervous system should work the same way. The context layer should be subconscious:

It captures without asking. It learns from the exhaust of work that already happens: calls, tickets, commits, emails, documents, decisions, system events. Nobody fills out a form.
It maintains itself. It reconciles conflicting facts, retires outdated ones, notices when a decision has been superseded and keeps the model current. Forgetting matters as much as remembering.
It routes context proactively. It doesn't wait to be searched. It brings up the relevant history when someone opens the account, joins the project or starts the task, the way a reflex fires before you decide to move.
It governs itself. Permissions, privacy and provenance are built in, so people can trust what shows up and see where it came from.
It stays out of the way. The best measure of success is that people stop noticing it. They simply seem to know what they need to know.

This is why the productivity gain is different in kind, not just in size. Point solutions save time on tasks. A subconscious context layer removes the invisible work around tasks: searching, re-explaining, status meetings, rework and "wait, I didn't know that." That surrounding work is often larger than the task itself.

What changes for people

The goal is not replacing people. The goal is freeing their conscious attention.

Today, a large share of knowledge work is really context work: finding, reconstructing and passing along what the organization already knows. When the nervous system takes over that load, humans can focus on what the conscious mind is for: judgment, tradeoffs, relationships, taste and deciding what matters.

Managers stop being human routers of information and become coaches and decision makers. New hires ramp in weeks instead of months because the company's memory is available to them from day one. Departures stop taking years of context out the door with them.

The thesis

AI has been deployed as point solutions that optimize functions, and those gains are real but local.
Company performance is limited by flows between functions, and the main constraint in those flows is context loss.
Organizational sync is the variable that governs those flows, and it can be measured on a maturity ladder.
Getting past the Augmented rung requires a company brain: a shared context layer under every tool and person.
That layer only works if it is autonomous and subconscious, because every knowledge system that depended on human upkeep has decayed.
Companies that build this nervous system will compound. Companies that keep stacking point solutions will get faster at being out of sync.

The next era of enterprise AI will not be decided by which company has the smartest tools. It will be decided by which company has the best nervous system connecting them.

Would love to hear what people think of these ramblings, the ideas around organizational sync and if a nervous system is the best way to think about the subconscious contextual knowledge management layer.