The wrong question
Most enterprise AI conversations start with the same question: how do we make this process faster?
That's automation thinking. It's useful, but it's not what AI actually makes possible — and it's definitely not what will separate the enterprises that scale from the ones that stall.
The better question isn't "how do we speed up the existing steps?" It's "if we were designing this from the outcome backwards, would we even have these steps?"
The analogy that stuck with me: digital transformation was about preparing the car for faster highways. Better roads, better engines, GPS instead of paper maps. The car was still the car.
AI is about flying. And flying runs on completely different physics.
If you keep optimising the car — adding automation, bolting on AI assistants, speeding up existing workflows — you'll build a very fast car. But you won't get off the ground.
"Most enterprises are asking how to make their processes faster. The real question is whether those processes deserve to exist at all."
1. First-principles thinking, not automation thinking
The trap is keeping humans in the loop in the wrong places.
When an agent does the research, surfaces the options, and then hands off to a human — not for a decision, but to second-guess the output and re-enter it somewhere else — you haven't built an intelligent workflow. You've built an expensive one.
First-principles thinking starts from the outcome. What needs to happen? What decision needs to be made? What action needs to be triggered? Then work backwards. If a step exists only because it existed before AI, it's a candidate for elimination — not optimisation.
This is harder than it sounds. Organisations are built around their processes. Questioning the process feels like questioning the people who run it. But the enterprises that will scale AI aren't the ones that automate everything that exists — they're the ones willing to redesign from the output backwards.
2. Systems of record are context-poor
ERPs and CRMs are excellent at capturing transaction state — the what. Order placed. Invoice raised. Case closed.
They're poor at capturing context — the why. Why was this supplier chosen over a cheaper alternative? Why did this deal close when three similar ones didn't? Why does this approval take three weeks when the policy says five days?
The "why" lives elsewhere. In emails, call logs, documents, approval threads, event streams. It's unstructured, distributed, and almost never connected to the transaction it explains.
This is the gap that breaks enterprise AI in production. Models trained or operating on transaction state alone make technically correct but contextually wrong decisions. They see the what. They can't see the why.
Closing this gap — building what I'd call an Enterprise Context Layer that captures, structures, and makes available the "why" alongside the "what" — is the infrastructure work that makes the difference between AI that performs in a demo and AI that performs in production.
3. Invented, not discovered
There's a distinction worth holding onto.
Fire was discovered. Our ancestors couldn't conceive where it would lead — steam engines, industrialisation, the modern world — because it was a force they stumbled upon and learned to use over centuries.
AI was invented. We are the architects. That gives us something our ancestors didn't have: foresight. We can, to some degree, design the trajectory.
That's both the opportunity and the responsibility. We're not waiting to discover what AI becomes. We're deciding what we build it into.
Deepak's Take
The leap from digital to AI is as significant as the leap from driving to flying. But that doesn't mean we need to build a spaceship on day one.
The path forward is disruption through evolution.
Start by augmenting existing systems of record with an Enterprise Context Layer — the layer that gives agents the business context, guardrails, and decision authority they need to operate alongside humans. As agents multiply and take on more of the operational load, the architecture matures. The augmented car becomes the flying car. The flying car, in time, becomes whatever the business actually needs.
The question isn't evolution versus reimagination. It's sequencing. Reimagination is the destination. Evolution is how you survive long enough to get there.
Deepak Choithramani is Co-Founder of Vishleshan AI Solutions. He writes about enterprise AI, agentic systems, and what it actually takes to go from pilot to production.
Follow Deepak on LinkedIn