Can the Institution Move as Fast as the Intelligence?

On day one of the 2026 Intelligence & National Security Summit, held just outside Washington, the question was no longer simply whether the intelligence community should use artificial intelligence. Faced with too much data, incompatible systems and technology evolving faster than government processes, the more urgent question was whether the institution itself can move fast enough.

Two words surfaced again and again on day one of the Intelligence & National Security Summit, held just outside Washington: speed and scale.

But beneath them was another word that may better describe the immediate obstacle confronting the intelligence community:

Interoperability.

The United States does not suffer from a shortage of information. Quite the opposite. Sensors, satellites, open-source intelligence, commercial databases, facial-recognition systems, cyber intelligence and countless other streams generate extraordinary quantities of data.

The problem is increasingly one of oversaturation.

There is too much information for human beings to process conventionally, spread across too many systems that often cannot communicate easily with one another.

Artificial intelligence offers a way to process that information at machine speed. But AI alone cannot solve the underlying architectural problem.

The information has to move.

The databases have to connect.

The platforms have to communicate.

The intelligence has to reach the right person at the right moment in a usable form.

And that may be the real choke point.

The Interoperability Problem

Throughout the summit, speakers repeatedly returned to the difficulty of connecting systems.

Greg Ryckman, Special Advisor to the Director of the Defense Intelligence Agency, put it plainly: solutions have to be “interoperable.”

The problem is easy to underestimate because the individual technologies can be extraordinarily sophisticated.

One system may collect information brilliantly. Another may analyze it. Another may visualize it. Another may support targeting or decision-making.

But intelligence is not created by possessing a collection of impressive technologies.

Those technologies have to work together.

A system that produces valuable information but cannot effectively pass it to another system creates another silo. A data set unavailable when an analyst needs it has limited operational value. An AI model isolated from the information required to understand a problem cannot compensate for the architecture around it.

The challenge therefore isn't simply building a better machine.

It is connecting machines, databases, organizations and people into a functioning whole.

The emerging intelligence architecture increasingly looks something like this:

sensors → data → AI processing → analyst → decision → action

But each arrow represents a potential failure point.

Can the data move?

Can one platform read what another produces?

Can information cross organizational boundaries?

Can analysts see the provenance of the information?

Can decision-makers receive it quickly enough to act?

The intelligence community increasingly possesses extraordinary individual capabilities. The harder task is making those capabilities behave like a system.

Drowning in Data

For much of the history of intelligence, one of the central challenges was acquiring information an adversary wanted to conceal.

That remains important.

But the digital environment has introduced almost the opposite problem.

There is now so much potentially relevant information that simply collecting more of it can make the intelligence problem worse.

More information does not automatically produce more understanding.

It can produce more noise.

The scarce commodities become attention, verification, provenance, interoperability, judgment and time.

Artificial intelligence becomes essential partly because the quantity of information has exceeded what conventional human workflows can reasonably process.

Asked how much should be automated, Ryckman's answer was striking:

“As much as possible.”

The logic appeared repeatedly throughout the summit: if intelligence organizations cannot automate substantial portions of information processing, they cannot operate at the speed and scale of the environment they are trying to understand.

“If you aren't automating, you can't keep up,” I recorded in my notes.

That is a very different proposition from using AI because it is novel or efficient.

Automation becomes a response to information abundance.

When Warfare Doesn't Look Like Warfare

The problem of data oversaturation becomes still more difficult because modern adversary activity does not remain neatly confined to the military domain.

One of the recurring concerns on day one of the summit was asymmetric campaigns—activity that may cross economic, diplomatic, informational, technological, cyber and military domains without initially presenting itself as warfare at all.

That distinction matters.

An intelligence system designed primarily to identify conventional military indicators knows what it is looking for: troop movements, weapons systems, aircraft, ships, launches and other observable manifestations of military power.

An asymmetric campaign can be considerably harder to recognize.

A commercial transaction may be only a commercial transaction. An infrastructure investment may be ordinary investment. A social-media campaign may be political speech. A cyber intrusion may appear isolated. Diplomatic pressure may be diplomacy.

But occasionally those events are not independent.

They are pieces of a larger campaign.

The intelligence challenge is therefore increasingly not simply detecting an event, but recognizing relationships among events occurring in different domains—and identifying a coordinated adversary strategy before its individual components become obviously connected.

That helps explain the urgency surrounding interoperability.

If economic intelligence sits in one system, cyber information in another, military intelligence in another, open-source information somewhere else and commercially collected data on yet another platform, an analyst may never see the pattern created by all of them together.

The technical problem of getting platforms to “talk to each other” therefore has strategic consequences.

Interoperability is what makes cross-domain pattern recognition possible.

And artificial intelligence potentially becomes especially powerful here. Machines can examine enormous quantities of heterogeneous information simultaneously, looking for relationships that no individual analyst could reasonably discover manually.

But this also raises the difficulty of the analytic task.

Finding a tank is one thing.

Determining whether a collection of financial transactions, information operations, commercial relationships, cyber activity and diplomatic behavior represents a coordinated adversary campaign is something else entirely.

The intelligence community is increasingly being asked not merely:

What is the adversary doing?

But:

Are these seemingly unrelated things actually connected?

And ultimately:

What does the adversary intend to do next?

One of my notes from the discussion captured the ambition starkly: “Can’t warn before an adversary's decision is made.”

That pushes intelligence further upstream—from observing actions toward recognizing campaigns and anticipating intent.

It also makes the problem of data oversaturation more acute. The relevant signal may already exist somewhere inside the enormous quantity of information being collected. The difficulty is recognizing which fragments matter, connecting them across domains and platforms, and doing so early enough for the resulting intelligence to be actionable.

In that sense, the intelligence problem is no longer simply finding the needle in the haystack.

It is recognizing that several needles scattered across several different haystacks belong to the same weapon.

Have the Machine Show Its Homework

Yet the enthusiasm for automation was accompanied by an equally persistent concern: trust.

Ryckman argued that AI needs to be as transparent as a human analyst.

Another phrase captured the idea even better:

Have the machine show its homework.

The two principles belong together.

Automate as much as possible.

But:

Make the automation legible.

If an AI system helps determine which pieces of information receive human attention, identifies relationships among disparate data sets, prioritizes threats or contributes to targeting decisions, the people relying upon it need some ability to understand how the machine arrived there.

That means provenance becomes more important, not less.

Where did the information originate?

What sources contributed to the conclusion?

What assumptions did the system make?

What information might be missing?

What level of confidence should an analyst assign to the result?

At machine speed, a mistake can also travel at machine speed.

The faster intelligence becomes, the more important the ability to interrogate its origins.

What Is an Analyst in the AI Era?

One of the most revealing questions I heard at the summit was also one of the simplest:

What is the definition of an analyst in an AI era?

The question matters because AI is beginning to participate in activities that once constituted much of analytic work.

Traditionally, an analyst might find information, read it, compare sources, organize evidence, identify patterns and produce an assessment.

Increasingly, machines can assist with almost every one of those steps.

That does not necessarily eliminate the analyst.

It may move the analyst to a different place in the process.

The emerging workflow could increasingly become:

The machine finds. The machine filters. The machine synthesizes. The human interrogates. The human judges.

That places a premium on distinctly human capabilities: skepticism, context, source evaluation, intent inference, accountability and judgment.

The phrase “human reserved jobs” surfaced during the discussion.

It may be a better framework than asking which jobs AI will eliminate.

The intelligence community will increasingly have to determine which activities can be automated and then make a separate judgment about which activities should remain human.

There is another danger.

If analysts surrender too much of the underlying work, they risk losing the expertise required to recognize when the machine is wrong.

The future analyst may therefore face an unusual requirement: become proficient enough with AI to exploit its speed while remaining sufficiently independent of it to challenge its conclusions.

The machine has to show its homework.

The human still has to know enough to grade it.

The Generational Hierarchy Is Beginning to Invert

There was another pattern at the summit that deserves more attention.

In institutions traditionally structured around rank, seniority and experience, technological expertise can invert the normal hierarchy of knowledge.

Senior military and intelligence leaders may possess decades of operational and institutional experience while younger service members have grown up inside the technological environment those institutions are now trying to understand.

That means expertise does not always flow downward.

Increasingly, senior leaders are looking down the chain of command for technological understanding—asking younger service members what they are using, what is changing, what works and what possibilities senior leaders may not even know to request.

That is more significant than it sounds.

The people with the authority to transform an institution are not necessarily the people most familiar with the technologies transforming it.

One speaker invoked a famous line commonly attributed to Henry Ford:

“If I had asked people what they wanted, they would have said faster horses.”

The quotation is widely attributed to Ford, although there is no reliable evidence that he actually said it.

But the idea behind it fit the conversation remarkably well.

Sometimes an institution cannot simply ask how technology can improve an existing process.

It has to reconsider the process itself.

The intelligence community does not merely need a faster horse.

It may need a different vehicle.

Looking Outside the Government

That helps explain another striking feature of the summit: the extent to which government appears to be looking outside itself for technological capability.

The private sector did not feel peripheral to the conversation.

It felt increasingly structural.

Commercial data providers, artificial-intelligence companies, cloud and computing infrastructure, software platforms and other private technologies are becoming intertwined with the government's ability to collect, process and understand information.

One line in my notes reads simply:

Government needs to keep up with civilian technology.

That may be one of the more consequential admissions of the entire conference.

Historically, some of the world's most advanced technologies emerged from military and government research and later migrated into civilian life.

Increasingly, important parts of the relationship run in the opposite direction.

Commercial technology can develop faster than government procurement systems can buy it, approve it or integrate it.

The emerging national-security architecture therefore looks less like a self-contained government apparatus and more like an ecosystem:

government ↔ private industry ↔ data providers ↔ AI systems ↔ allies ↔ operators

That relationship offers enormous advantages. Government can access innovation occurring across a technological economy vastly larger than any single agency.

But it creates another set of questions.

What happens when critical government capability depends on proprietary commercial systems?

Who controls the underlying technology?

Can different vendors' platforms communicate?

Can the government move data between them?

Can an agency replace one provider without rebuilding the architecture around it?

And what happens when the private sector continues evolving faster than the government's ability to integrate what it produces?

Interoperability therefore isn't merely a technical problem.

It is an institutional and strategic one.

An “I'll Try Anything” Moment

The atmosphere captured in my notes was not one of an institution calmly selecting among mature technological options.

It felt closer to an “I'll try anything” moment.

There was urgency.

The existing systems are too slow. The information environment is too large. Adversary activity does not remain conveniently inside military categories. Technology is changing too quickly. Government processes have not caught up.

That creates a peculiar historical moment in which experimentation becomes not merely desirable but necessary.

AI.

Automation.

Commercial data.

New sensors.

Private-sector platforms.

Different workforce models.

New relationships between junior technological expertise and senior institutional authority.

The intelligence community appears willing to experiment across all of them because the alternative—continuing to process an exponentially more complicated world through structures designed for a slower one—is increasingly untenable.

That willingness carries risks.

An institution under pressure to move faster can adopt technology before fully understanding its consequences.

But excessive caution creates a different risk: irrelevance.

The tension throughout the summit was therefore not simply between humans and machines.

It was between institutional caution and technological acceleration.

The Real AI Threshold

I arrived at the summit wondering whether the intelligence community had crossed an AI threshold.

I think it has.

But the threshold may not be primarily technological.

It is organizational.

The critical moment arrives when machines become capable enough that the institution surrounding them must change.

Once AI can search, triage, connect and synthesize information at enormous scale, human-speed processes become increasingly visible as bottlenecks.

Then the questions change.

Not:

Should we use AI?

But:

How much can we automate?

How do we connect the systems?

How do we make the data move?

How do we know where the information came from?

What decisions remain human?

How do we keep the workforce technologically relevant?

How does government absorb innovation occurring outside government?

And ultimately:

How do we redesign an institution whose underlying assumptions were created before the technology it now depends upon existed?

That produces a series of paradoxes.

More data requires more filtering.

More automation requires more transparency.

More speed makes judgment more valuable.

More powerful AI makes provenance more important.

More technological capability makes interoperability more urgent.

More reliance on machines makes the remaining human decisions more consequential.

And greater dependence on commercial innovation makes the boundary between public and private national-security capability increasingly difficult to define.

The intelligence community does not appear to be suffering from a shortage of intelligence technologies.

It is struggling to connect them—to one another, to the data, to analysts and ultimately to decisions.

The intelligence can move faster.

The technology can move faster.

The private sector is already moving faster.

The question now is whether the institution connecting all of them can move fast enough to matter.

Author's AI Disclosure

This article is AI-written from human-generated source material and direction. The underlying reporting, observations and handwritten conference notes were produced by Ashley Greer on day one of the 2026 Intelligence & National Security Summit. Ashley subsequently used conversational prompts to identify recurring themes, test interpretations, establish the article's argument and direct revisions. AI was used to transcribe and organize portions of the notes, analyze recurring themes and write the article's prose. The reporting observations, source material, editorial direction and final responsibility for the article remain with the author.