Agentic bloat: Why collecting AI tools won’t save you

The pattern we’re seeing

Across professional services, we keep meeting leaders who’ve done what feels like the responsible thing with AI. A research tool for one team. A proposal generator from one vendor, a meeting summarizer from another. Maybe an agent that pulls project data, and a separate one that drafts status reports.

Each purchase makes sense on its own. Each solves a real problem. Together, they quietly recreate the same fragmented knowledge problem—and the same operational overhead—that made AI necessary in the first place.

A growing number of firms are trying to fix this by building their own connective layer instead. But what exactly are they building, and what happens when the person who built it moves on?

Gartner projects that more than 40% of agentic AI projects will be cancelled by the end of 2027. That’s not a failure of ambition. It’s the predictable result of treating AI adoption as a procurement exercise instead of an architectural one.

Why it feels right, but isn’t

The individual tools aren’t the problem. They work. A meeting summarizer does summarize meetings. A proposal generator does generate proposals. The efficiency gains are real, if modest.

The problem is that none of them talk to each other. They can’t share context, and they don’t know what the others know.

A meeting summarizer captures a client’s concern about timeline risk. A proposal generator, running on entirely separate context, produces a scope that doesn’t account for it. A status agent reports metrics that look fine because it has no visibility into the sentiment data sitting in another tool’s silo. Layered on top of that, the people prompting these tools often still have the same context gaps they had before AI—the information never reached them either.

The result is a set of AI islands: the same knowledge silos that have shaped professional services for decades, now running on AI instead of email and spreadsheets, and no more capable of compounding than they ever were.

Recent research puts a number on it: 70% of enterprises haven’t moved past basic integration for their AI tools, and three in four have experienced at least one negative outcome from disconnected AI systems. This isn’t early-adopter friction. It’s structural—and our own research bears it out. In the 2026 Global Service Dynamics Report, 98% of professional services organizations report having deployed some form of AI, but only 20% say results have exceeded expectations.

When agents add work instead of removing it

It gets worse before it gets better. Deloitte’s recent analysis on agentic AI describes “workslop”: agentic applications that add work to a process rather than remove it.

This happens when agents operate without shared context. Their outputs need human reconciliation. Someone has to cross-reference the meeting summary against the proposal. Someone has to check the project status against what was actually said on the client call. Someone becomes the integration layer the technology was supposed to provide.

Deloitte also flags “agent washing”—vendors rebranding existing automation as “agents” to ride the hype cycle. A lot of what gets called agentic AI is really automation wearing a new label, applied where a simpler tool would do.

Put together, this is the real cost: paying for the appearance of AI transformation while creating new reconciliation work that didn’t exist before.

The root problem: context without architecture

Collecting AI tools is a symptom. The disease is that most firms’ institutional knowledge has no coherent architecture. It’s scattered across:

• Individual consultants’ heads—and it leaves when they do
• Project folders organized by engagement, not by insight
• Chat threads that disappear into the void
• Meeting recordings nobody has time to watch
• Methodologies buried in documents nobody can find

Bolting more AI tools onto that doesn’t fix it. It accelerates it. Every agent is only as intelligent as the context it can reach—fragmented context produces fragmented output, confident-sounding mess at scale.

“We’ll just connect everything”

The usual next move is an integration play: pipe data between tools, or pour everything into a data lake with dashboards on top.

It sounds reasonable. It misses the point.

Connectors move data. They don’t create understanding. Syncing meeting transcripts to a CRM and a project system doesn’t mean a proposal agent understands the client relationship—it just means the same fragmented information is stored in more places. A data lake full of unstructured context is still a lake. Agents can swim in it. They can’t reason against it.

The difference between integration and architecture is the difference between having every ingredient in the same kitchen and having a recipe. One is a prerequisite. The other actually produces something.

The foundation that has to come first

Before adopting another AI tool, three questions are worth answering honestly:

Where does collective intelligence  actually live? Not where it should live—where it actually lives, in what systems, in whose heads, in which formats.

How would an agent access it? If the answer involves querying seven systems with different permissions and no common schema, that’s the real problem to solve.

How does learning compound? When a team learns something on one engagement, does that insight reach the next one—or does it depend on someone remembering to mention it? If it’s the latter, that’s a knowledge lottery, not an architecture.

This is what a unified foundation means in practice: a governed layer that lets AI systems access, understand, and build on a firm’s collective intelligence, instead of operating in silos next to it.

Built in the right order

The firms that come out ahead in this transition won’t be the ones with the most tools. They’ll be the ones that built in the right order:

First, a connected foundation—the record that ties sales, delivery, customer success, and finance together, along with the calls, notes, and messages around them. This isn’t a six-month side project; it’s the strategic decision everything else depends on.

Second, selective tool and agent adoption, evaluated against one question: does this compound on the foundation, or does it create another island? Anything that can’t plug into the architecture isn’t an investment—it’s future technical debt.

Third, orchestration over collection. The goal was never the most agents. It’s agents that share context, learn from each other’s outputs, and make the firm smarter over time. One well-orchestrated system beats ten disconnected point solutions.

This cycle moves faster than the ones before it. Tools are easier to adopt, silos multiply quicker, and the reconciliation work piles up before anyone notices they’ve created it.

The right question isn’t “which AI tools should we buy?” It’s “what does our foundation need to look like for AI to actually compound what we already know?” That’s a harder question—and it’s the one build vs. buy really comes down to, since neither choice matters much until the context problem is solved first.

See how Certinia’s System of Action gives every Veda agent one governed source of context to work from. Talk to our team.