The pitch is that AI agents do the work. The part nobody puts on the slide: an agent is only as good as the data it can reach. When agents make building cheap and fast, the bottleneck doesn't vanish — it moves up to two things that are entirely yours: knowing exactly what to build, and having clean, reachable data to build it on. Get those right and agents compound. Skip them and you've just bought faster chaos. This is that argument, in plain terms.
Buying AI agents before your data is ready is like hiring a brilliant team and giving them a filing cabinet full of unlabelled, half-empty folders. They'll work fast — and confidently produce wrong answers. The winning move isn't "deploy agents." It's fix the data layer and sharpen what you ask for first, then let agents run on top. The readiness work is the investment; the agents are the payoff that follows it.
The headline everyone hears is "agents will replace people." The more useful truth, from Andrew Ng and Harrison Chase at LangChain's agent conference, is quieter: when building gets cheap, whatever sits above building becomes the new constraint.
"Agents do the work, so the engineer is the thing we no longer need." Attention goes to the tool, the demo, the headcount you'll save.
Building is cheap now, so the limit moves up to scoping (knowing what to build) and data (something accurate for the agent to stand on). Those are management problems, not engineering ones.
Both of these used to be hidden behind the slow, expensive build step. Now that building is fast, they're exposed — and they're squarely the leadership's job, not the technology's.
An agent that builds the wrong thing in an afternoon is worse than an engineer who builds the right thing in three weeks. Precise scoping — clear requirements, clear boundaries for what a "digital worker" may do — becomes the rate-limiting skill.
Most businesses keep their real information scattered across silos, half of it unstructured. Agents need continuous, structured access to the actual state of the business. No clean data layer, no trustworthy agent — full stop.
This is the single most expensive misunderstanding about agents. They don't fail loudly. They fail plausibly — producing a confident, well-written answer built on data that was stale, missing, or scattered.
The data an agent needs to stand on — your records, your suppliers, your finances, the real state of the business — is exactly the back office below the waterline. Agents are the visible tip; the owned, reachable data layer is the mass that holds them up. That's why this is the same argument as owning your stack, seen from the other end: you cannot ground an agent on a black box you don't control.
Most first AI projects aim at trimming cost — faster admin, quicker approvals. Useful, but it's the small prize, and it's where the thinking usually stops.
Shave time off existing tasks — faster underwriting, quicker reports. Real, but capped: you can only cut cost down to zero.
New services, new markets, capabilities you couldn't offer before — built by combining reusable agent building blocks. Top-line growth has no ceiling, and it compounds.
The technology shifts monthly. Build on open, modular pieces — open-source models and frameworks — so you can swap in a better or cheaper model without rebuilding. Don't let your data-and-agent layer get locked to one vendor at exactly the moment the field is moving fastest. Owning the back office is what keeps that door open.
"Are we AI-ready?" isn't answered by buying software. It's answered by an honest look at your data and your scoping discipline — and then fixing them.
This briefing makes the case. Doing it — the AI-readiness audit (how clean and reachable is your data, really?), the data rearchitecture that gives agents something solid to stand on, defining what your digital workers may and may not do, and — in a South African context — deciding where that data lives under POPIA — is the work Imbila does. It's the gating investment that turns "we should use AI" into AI you can actually trust.
Imbila — AI strategy that delivers ROI ↗Before any agent project: how clean, complete and reachable is the data it would rely on? Honest answer first — this is usually the real work.
Decide precisely what you want built and what a digital worker is allowed to do. Faster building only helps if you're building the right thing.
With clean data and clear scope, put agents on top — on open, swappable foundations — and let the capability compound.
Don't buy agents and hope. Fix the data they stand on and sharpen what you ask for — then the agents are almost all upside.