top of page
Insights
Search

How to Build the Superintelligent Enterprise

  • Writer: Ram Srinivasan
    Ram Srinivasan
  • Jul 8
  • 6 min read

IA FORUM MEMBER INSIGHTS: ARTICLE


By Ram Srinivasan, Managing Director, AI Adoption & Integration & Future of Work Advisory, JLL

 

Matthew Gallagher, a self-taught coder, built a company that generated $401 million in revenue in its first full year, 2025, with a core team of two: him and his brother. They started with about $20,000 and more than a dozen AI tools, plugged into an existing telehealth supply chain. Today the business is reported to be on track for roughly $1.8 billion in annual revenue and a run rate of nearly $5 million a day.

 

That should make every large company uncomfortable.

 

Enterprises have more data, more customers, more expertise, and more capital. They also have scattered systems, slow committees, fragile approvals, and operating models built around handoffs.

 

The question is whether they can turn AI from a desktop tool into an operating capability.

 

A superintelligent enterprise is a company where intelligence sits inside the workflow. Agents can reach the right context, draw on trustworthy data, use approved tools, remember what happened, continue to work overtime, and return to humans at clear review points.

 

Microsoft's Satya Nadella published an essay recently titled "A Frontier Without an Ecosystem is Not Stable". As Satya notes, a firm now holds two kinds of human capital and token capital:

 

  • Human capital is the ingenuity, knowledge, relationships, context, and pattern recognition of an organization's people.

  • Token capital is the specific, AI capabilities, workflows, internal agents, and digital memory that a company builds and controls.

 

The work is to compound the two in a learning loop that encodes the company's own knowledge, instead of renting intelligence from a model that keeps everything it learns. The failure he names is the company that delegates the work but never captures the learning, feeding someone else's system instead of its own.

 

Seven Moves to Build the Superintelligent Enterprise

I believe going from a single superintelligent user to superintelligent team to superintelligent enterprise requires seven moves:

 

1. Start with Vision

A loop needs a destination. Vision is the concrete picture of what the enterprise is building toward. Not a mission statement, but a definition of done at the company level: what the work is for, what better looks like, and how the company will know if it got there. Without it there is no purpose. You get a drawer full of pilots, each defensible on its own and pointless together. You may get a lot of great energy, but it is directionless.

 

2. Connect the Context

An agent is only useful if it can reach the work. In most companies the work is scattered across CRM, ticket queues, repositories, email, shared drives, contracts, and finance tools. The first job is to make those systems reachable, governed, and current, so an agent can pull the right state, at the moment it acts. Many AI failures are context failures: the model cannot reason over what it cannot retrieve.


3. Build the Data Foundation 

Reach is not enough if what the agent reaches is wrong. Underlying the systems is the data itself, and it has to be trustworthy: a clear source of truth, clean, current, with lineage you can check. Where does the truth live? How fresh is it? How does the company know an answer came from the right place? Without a foundation, the agent does not fail loudly. It guesses confidently, which is worse.

 

4. Give the Work a Loop

The old unit of work was the task. The new unit is the loop. A loop watches for inputs, retrieves context, calls tools, drafts work, updates systems, keeps state, escalates exceptions, and returns to a human for review. A chatbot that answers once is useful. A loop that monitors a ticket queue for a week, drafts responses, updates the CRM, routes challenging cases, and learns from review is infrastructure. As model costs fall and autonomy improves, advantage shifts from having access to AI to designing the workflows around it.

 

One note here is that we build the loop knowing the process around it will have to change. Otherwise, you risk automating the old job i.e., paving the cow path. Instead, what we are doing is learning what the new one should be.

 

5. Govern the Agents

Once an agent can act, governance has to be built into the system. A wrong answer wastes time. A wrong action can move money, email customers, expose data, or push broken code into production. Every agent needs clear boundaries: What can it read? What can it write? What can it send? What can it approve? What requires human signoff? Who owns it? Where is the audit trail? How can it be shut down quickly? This policy cannot live in a document nobody reads. Legal, Risk, Compliance, Security, HR, and business leaders need one operating model, enforced through identity, permissions, logs, approvals, and kill switches.

 

6. Redesign the Work

Adding agents to old processes usually produces small savings. Larger gains come from redesigning the process itself. The useful question is simple: if drafting, routing, retrieval, reconciliation, summarization, and first-pass analysis become cheap, how should the work be organized? That question changes roles, teams, approvals, metrics, and management. The progression is predictable: the first agents automate tasks, the next wave reshapes jobs, the largest wave changes the org chart. This is why I have maintained every AI conversation is an operating model conversation.

 

7. Keep the Human Seat

A loop has no taste. It scales the judgment pointed at it. That makes human judgment more important, especially at two points: the goal at the front and the review at the back. The goal must be specific enough for the loop to test its work against it. The review must be real for anything involving customers, money, legal exposure, safety, reputation, or production systems. Anthropic, Google, OpenAI, and others have already said a large share of new code is machine-written. The bottleneck has moved from writing to reviewing. That pattern will spread across the enterprise. When execution gets cheaper, judgment becomes more valuable. Put the strongest people where goals are set and where important work is reviewed. Let the loops carry more of the middle.

 

Why this Matters to You

The superintelligence enterprise is already forming at the edge.

 

For example, Claude is entering Slack as a persistent teammate, with Anthropic’s Claude Tag allowing teams to treat it as a governed, context‑aware agent in their channels rather than just a chatbot. And Microsoft and Google have moved on horizontal agent platforms aimed at building, running, and governing fleets of agents across an enterprise, not just measuring human work.

 

Here’s your move:

 

  • If you run the company, shift budget from pilots to workflow redesign.

  • If you lead transformation, own the context layer, the data foundation, the loops, and the review gates.

  • If you own risk, treat action-taking agents as both a new attack surface and a new control surface.

  • If you manage people, prepare each person to supervise a portfolio of loops.

  • If you are early in your career, build skill in goal setting, review, domain judgment, and taste.


The marginal cost of intelligence keeps falling. The length of work AI can sustain keeps rising. The constraint moves to the two things a machine cannot supply: the vision that says what is worth building, and the judgment that says when it is built. The enterprises that win will get those two right, compound their human and token capital in loops they own, and point them at everything in between.

 

None of this is automation for its own sake.

 

The whole structure exists to put people in charge of more than they could hold before. It should take human judgment, taste, and accountability and give them reach. A superintelligent enterprise is not one that runs without people. Quite the opposite. It is one designed so that human agency sits at the center, with the machines arranged around it.

 

Build it human by design.

 

Author Disclaimer: The views and opinions expressed herein are those of the Author alone and are shared in a personal capacity, in accordance with the Chatham House Rule. They do not reflect the official views or positions of the Author’s employer, organization, or any affiliated entity.

 

References

Comments

Rated 0 out of 5 stars.
No ratings yet
Add a rating
bottom of page