Synergised Consulting
Technical asset

Bolt the Agent On or Rebuild the Workflow? How to Tell Which One Yours Needs

Synergised Consulting Ltd

5 min read
Title card: Bolt the agent on or rebuild the workflow, with the line that only 1 in 5 leaders say they are ready to redesign

Last Updated: 13 November 2026

Most AI agents underdeliver because they are bolted onto a process built for human hand-offs, yet rebuilding every process is neither affordable nor necessary; the practical answer is to test each workflow and let its own shape decide the treatment. That decision matters now because adoption is running well ahead of redesign. Deloitte's 2026 survey of 501 US leaders at organisations already piloting agents found 43% deploying agents across multiple functions, but only one in five saying their organisation is prepared to redesign processes for autonomous operation [1]. The gap between those two numbers is where most agent programmes quietly lose their value.

What bolting an agent on actually produces

A bolted-on agent inherits a process that was never designed for it, and the seams show. Picture an invoice approval chain: a purchase order arrives by email, someone matches it to a delivery note, a manager approves anything over a threshold, and the accounts system gets updated by hand at the end of the week. Drop an agent in to "automate the matching" and it now waits on the same email format, hands back to the same manager with the same context-free question, and still ends its run at a manual step designed for a person's Friday afternoon. The work got faster in the middle and stayed exactly as slow everywhere else.

That pattern is not hypothetical, it is the measurable norm. In the Deloitte research, most of the scaling that is happening stays in lower-risk, repeatable use cases such as customer service, IT and engineering [1], and the survey's own framing of the common pattern is layering agents onto existing processes in pursuit of quick returns. Only 5% of organisations describe their business processes as highly prepared for agents at all [1]. Bolt-on is not wrong; it is just the entry tier, and it produces exactly what the entry tier produces: a faster first draft, sitting inside an unchanged process.

The adoption numbers behind the rebuild question

The organisations scaling agents furthest are the ones redesigning, and very few are there. Deloitte's survey, fielded between April and June 2026 across 501 senior leaders, found 43% expanding agent deployments across functions but only 15% at scaled, orchestrated, multi-agent adoption [1]. Expectations run far ahead of that: 74% of leaders expect nearly half of their business processes to be redesigned or rebuilt around agents within four years [1]. Meanwhile the same survey found only 5% of organisations rate their processes as highly prepared for agentic adoption [1].

Three numbers, one conclusion. Deployment is common, redesign readiness is rare, and the industry expects the rebuild anyway. The danger for a small business owner is reading the expectation line and concluding that a redesign programme is the price of entry. It is not, and the survey itself contains the reason: 75% of leaders agree that human collaboration with agents creates more value than agent-powered automation alone [1], which means the redesign that matters is the one that decides where the human stays in the process, not the one that removes them.

How to tell which treatment a workflow needs

The decision rule comes down to hand-offs. A workflow whose steps mostly pass between people, each waiting on the previous one's judgement, needs its sequence rebuilt before an agent can do more than type faster; a workflow that is one person carrying a repeatable end-to-end task can take a bolted-on agent and capture most of the gain on day one. Deloitte's finding that only one in five organisations is prepared to redesign for autonomous operation [1] is consistent with what we see in discovery workshops: the workflow map rarely arrives written down, and it gets built in the conversation, walking through a normal week task by task. It is usually in that walk that the hand-off count becomes obvious, and with it the treatment.

The one-workflow test, in practice

Pick one workflow and write down four things before any agent touches it: the steps as they actually happen, the hand-offs between people, how long one pass takes, and where the errors surface. Then run the test sheet against the decision rule. A three-step, single-owner task with no approvals comes out "bolt on", and the agent should enter where the work is most repetitive. A seven-step task with two approval gates and a weekly batch update comes out "redesign first", because those gates were designed for a person's attention, not a machine's, and bolting on simply moves the queue.

The test sheet itself, with the map, the baseline and the chosen treatment on one page, is worth keeping whatever the outcome. It is the record of why the workflow looks the way it now does, and it is the first thing worth showing anyone who asks how you knew the change was safe to make.

What the first 100 days version looks like

For a new owner this test costs a conversation and an afternoon, because you are already walking the workflows that matter. Run it on the one process whose slowness you have already noticed. If it comes out bolt-on, deploy the agent this week and measure the same numbers afterwards. If it comes out redesign, you have learned something more valuable than a speed gain: you have found the process whose design predates the business you are trying to build, and the rebuild is now a scoped decision rather than an open-ended programme.

Sources

  1. [1] Deloitte Insights, The path to agentic transformation, published 11 August 2026, survey of 501 US senior leaders fielded April to June 2026: