AI Is Pushing Composability Beyond Software

A while ago I was chatting with a friend who works in marketing technology. He showed me a screenshot of his company's tech stack.

A dense grid. Nothing but boxes.

He pointed at one — that's the CRM. Pointed at another — that's the CDP (Customer Data Platform). Another — the automation platform. Then he gave me a wry smile: every tool has to be bought, every tool has to be wired in, and once wired in you still need to keep a whole team on staff just to babysit them.

One word popped into my head: LEGO.

What does "composable" mean?

For the past decade-plus, the enterprise-software world has been obsessed with one word: composability.

In plain terms, it means taking a giant all-in-one software suite and breaking it into swappable, snap-together little bricks.

Want to swap out your CMS? Swap it. Want to bolt on a recommendation engine? Bolt it on. The day your CDP stops pulling its weight, you pull it out and slot in another — without tearing the whole stack down.

It doesn't sound sexy. But it genuinely changed one thing: companies stopped being held hostage by any single platform. Used to be, if you wanted a new capability, you begged the vendor for a slot on the roadmap and paid a fortune. Now you grab a brick off the shelf and click it into place.

How many bricks are on that shelf right now?

Fifteen thousand five hundred and five. That's the 2026 number.

But here's where it gets interesting.

When the LEGO bricks start to "think"

Sometime last year, I noticed a shift.

The bricks that used to "just follow orders" started thinking for themselves.

A research agent can run a full competitive analysis for you. A content agent can take a draft from rough to publish-ready. A decision agent can comb through a pile of data and tell you where to move budget next. They don't wait for your command — they reason, plan, and collaborate, running themselves toward a goal.

That's a qualitative shift.

The old LEGO bricks were inert — whatever you built, that's what they were. The new LEGO bricks have grown a brain.

One set of numbers genuinely rattled me: 90.3% of surveyed companies have already wedged AI agents into their own marketing stacks; on average each firm is experimenting with 6.67 different types of agents.

6.67 types. This isn't dabbling — this is assembling an entirely new kind of operation.

So what about the old SaaS? Is it about to get knocked over?

SaaS isn't dead — it became the foundation

A lot of people see how hard AI agents are charging and their first reaction is: so should we throw out the CRM, the CDP, all those relics?

I thought so too, at first.

But the data tells a different story.

85.4% of companies are using AI to augment their existing marketing-tech capabilities; only 30.1% are using AI to replace legacy features.

Why no wholesale replacement?

Because these two categories of things, fundamentally, do two different jobs.

SaaS is deterministic: how customer records are stored, how permissions are managed, how orders flow — these things must be reliable, repeatable, auditable. Get one ledger entry wrong and finance implodes.

AI is probabilistic: it reads context, reasons across information, then tells you "here's the best next move." What it hands you is judgment, not record.

SaaS remembers what the business already knows. AI figures out what the business should do next.

One holds the line. One scouts the road ahead. These have never been the same job.

So most of the architectures you see look like this: SaaS on the bottom as the foundation, agents on top as the brain. If the foundation is shaky, no matter how fast that upper brain spins, you're still building on sand.

But the real change is inside the organization

At this point you might think the story's over.

It isn't.

AI has pushed composability out of software and into the organization itself.

Think about it — if capabilities can be broken into modules, and agents can work like colleagues, why does a team still have to be shaped like "Marketing, Sales, Operations"?

The marketing leader of the future isn't holding a roster of fixed headcount. She's holding a recomposable set of capabilities: human specialists, SaaS platforms, AI agents — three categories laid out together, assembled on demand around whatever business problem needs solving.

Whoever's best at something takes that piece. Humans are good at empathy and judgment. Agents are good at scale and reasoning. SaaS is good at storing data and running process.

This isn't science fiction. The marketing-ops role is already quietly shifting: it used to be administrators managing tools; then it became onboarding use cases; next it becomes engineers accounting for business value.

From babysitting tools, to designing capability combinations.

But there's a catch

Trust.

One report says only 6% of companies dare let agents run core business processes fully autonomously.

Another is more sobering: 80.6% of agents today still operate in "suggest, human decides" mode; those that can truly execute on their own and roll themselves back when something goes wrong — only 9.7%.

Why? Because when an agent goes wrong, things actually break.

Three in the morning. No one watching. An agent suddenly decides to double a client's promo budget and push it live. That image — whoever pictures it in the middle of the night breaks out in a cold sweat.

So for the foreseeable future, the mainstream architecture will sit at "AI does the work, humans supervise, humans make the call at the critical nodes." Some people call this the agent manager — in plain terms: you have to lead a fleet of AI the way you'd lead a team. Watching, evaluating, coaching, constantly tuning.

So where does this go next?

Look back at the thread:

Yesterday we were composing software. Today we're composing intelligence. Tomorrow what we'll compose is skills, organizations, and who does what.

The granularity of the LEGO keeps getting finer — from suites, to bricks, to thinking agents, to organizational forms that can redefine "who does what."

But one thing hasn't changed: human judgment is always the last gate in this system.

What AI genuinely changes is "what humans go do."

Next time you see one of those dense tech-stack screenshots, remember: that's not just an inventory of tools — that's an org chart being reassembled in real time.

Whoever makes sense of that picture first wins half a step ahead of everyone else.