Most organisations point AI at the process they already run. The ones pulling ahead are doing something harder, and far more valuable: redesigning the work itself.

The instinct that quietly caps your return

When a new capability arrives, the natural first move is to point it at the work in front of us. A finance team that loses three days a month to reconciliation asks how AI can make those three days faster. A logistics team buried in exception emails asks how AI can draft the replies. It feels responsible: start where the pain is, automate the obvious. And it does produce something - a process that runs a little quicker, a few hours handed back.

The problem is that it leaves most of the value on the table. You have made the existing process faster without asking whether the existing process should exist at all. Many of its steps are there only because a person used to do them: the handoffs, the approvals, the status-chasing, the re-keying between systems. Automate those faithfully and you have preserved the scaffolding of a way of working that was designed around human limitations, then taught a machine to climb it.

The mistake is where you start, not which tool you pick

This is not a technology problem. The tools are capable; the gap is in how the question is framed. “How do we automate this process?” and “how should this work be done now?” lead to completely different programmes. The first optimises a path that already exists. The second is willing to delete the path.

The question is not “what can we automate?” It is “how should this work be done, now that software can reason and act rather than only calculate?”

That distinction has become the clearest line between AI programmes that disappoint and ones that do not.

What the evidence is showing

The pattern is consistent across the major advisory firms writing on this in 2026. Deloitte’s work on agentic AI argues that leading organisations are not layering agents onto existing workflows; they redesign the work end to end and then manage the resulting agents as a kind of digital workforce, with defined roles and oversight. BCG has put numbers to the same point, reporting that the organisations capturing the largest cost reductions from agentic deployments are precisely those that redesign processes end to end, while those that bolt agents onto today’s process capture a fraction of the benefit. The framing has even acquired a name in the commentary: automating the past instead of designing for the future.

The throughline is simple. The differentiator is no longer access to the technology, because almost everyone now has that. It is the willingness to change the process the technology runs on.

What “redesign” actually looks like

Redesign sounds abstract until you see it on a real process, so here is the shape of it. Take any end-to-end process that today runs as a sequence: step one finishes and hands to step two, which finishes and hands to step three, and so on down the line. A surprising amount of the elapsed time is not work at all; it is waiting in the queues between steps.

When software can reason about context and act across systems, that sequence stops being necessary. Steps that used to wait for one another can run in parallel. Work that existed only to move information from one person to the next - the chasing, the re-entry, the “just confirming” emails - can be removed rather than reproduced. A process that ran one-to-ten in strict order becomes several overlapping workstreams, with people focused on judgement and exceptions while agents handle the connective tissue in between.

The honest version of this also means that some steps disappear and some roles change. That is uncomfortable, and it is also where the value sits. A redesign that leaves every step and every handoff intact is not a redesign; it is the old process with a faster engine bolted on.

Why this matters more in 2026 than it did a year ago

For most of the last few years, enterprise AI suggested. It surfaced an insight, drafted a paragraph, flagged an anomaly, and a person decided what to do next. That kept a human in every step, which is exactly why automating the existing process was good enough: the AI was an assistant working inside a human workflow.

That has changed. The platforms most large organisations already run have moved from suggesting to acting. On the SAP side, the Joule agent builder reached general availability in 2026, with agents designed to execute multi-step processes across finance, supply chain and HR. On the Google Cloud side, the Gemini Enterprise Agent Platform now offers a managed environment for building and governing agents that act on enterprise systems. Individual agents and features sit at different stages of release, some generally available and some still in preview, so the precise capability is worth checking at the time of reading.

Once software can carry a process rather than assist a person through it, redesigning the process stops being optional polish. It becomes the thing that unlocks the capability you have already paid for. And because production adoption is still early, with industry surveys in 2026 suggesting only a small minority of organisations are running these agents in live operations, the advantage still belongs to those that move deliberately now.

Where to start

Redesign is not a leap of faith; it is a method. We tend to work through four moves, one function at a time:

1. Map the function end to end and mark every step that exists only because a human has been doing it.

2. Reimagine the process for people and agents working together - turning sequences into parallel workstreams and removing the steps that were pure handoff.

3. Build the agents and the interfaces into the systems you already run, including the ERP, so the redesign works on your existing estate rather than something you have to replace.

4. Run and govern it, treating agents as a measured part of the workforce with clear oversight, escalation and accountability.

The order matters. Most disappointing programmes invert it: they buy or build the agent first, then go looking for a process to point it at. Starting with the function, and being willing to change it, is what turns AI from a faster version of yesterday into a better version of tomorrow.

The question worth asking

If there is one test to apply before any AI initiative, it is this: are we making an existing process quicker, or are we asking how this work should be done at all? The first is worth doing. The second is worth far more, and it is the one most organisations skip.

At VE3 we work alongside teams to do the harder, second thing: reimagining how a function actually operates, and connecting that redesign into the systems you already run. If you are weighing where AI could change not only the speed but the shape of your work, that is the conversation we would want to have.