Agent Desktop — the refresh nobody had funded
Data-sovereignty concerns were pulling enterprise interest back toward on-prem, and the agent’s daily surface was visibly ageing. But leadership’s attention sat with AI, so the work had no mandate, no budget and no team. Nobody was against it. It simply never rose.
Voice, digital and the customer record, on one surface. The live call sits beside the customer’s details and the chat that preceded it — a bot conversation about a delayed flight, the customer asking for a call, and the agent picking it up mid-thread. Transcript, suggested answers and the booking itself are tabs on the same pane rather than separate tools. This is the ambition the roadmap was arguing for: not a repaint, but voice and digital unified and CRM data brought into the agent’s workflow.
mandate, budget or team when the work started — and no one arguing against it either.
customers who raised the ageing experience independently, without being asked.
designer on it, working off the side of an already-full desk.
from decision to a prototype real enough to be mistaken for the product.
-
Diagnose the neglect correctly
Nobody had rejected the refresh. It was passive neglect, not opposition — which is the harder problem, because there is no argument to win, only invisibility to solve.
-
Wait for evidence you did not manufacture
Two customers flagged the ageing experience independently, unprompted. Advocacy you generate yourself reads as preference; the same point from outside the building reads as a signal.
-
Build the thing instead of pitching it
A working prototype went up in days, solo. Leadership was shown progress, not a proposal — which removes the step where they have to imagine it.
-
Borrow the design system’s credibility
It was built on the existing design system, so it read as the product rather than a concept. A prototype that looks speculative gets debated; one that looks shipped gets scheduled.
-
Sequence the commitment
The vision was split into a phased roadmap that could be approved a step at a time. The first yes was small enough to be easy, and every one after it had a precedent behind it.
-
Hand the method back
The outcome that mattered longest was not the refresh. Evidence over argument became something the wider team now reaches for by default, on work that has nothing to do with this one.
- Leadership attention sat with AI; the Agent Desktop never rose above competing priorities.
- No mandate, no budget, no team — the work had to be done off the side of a desk before it could be asked for.
- Passive neglect is harder than resistance: there is no counter-argument to answer, only invisibility.
- The need was genuine and externally driven, but it had no internal advocate until one was built.
The refresh moved from an exec-invisible priority to a committed initiative with a funded first phase, now in execution. The larger result was the method: a prototype built off the side of a desk, on the real design system, sequenced into approvable phases — a pattern the team repeats now without being told to.
Build it, don’t pitch it
A deck asks an executive to imagine an outcome and then fund the imagining. A working prototype removes that step entirely — what is being funded is already on the screen.
Make it read as real
Using the shipping design system rather than a concept style was the decision that did the persuading. It moved the conversation from should we to when.
Sequence the commitment
One large unfunded ask invites a no. A phased roadmap turns it into a series of small yeses, each carrying the evidence of the last one behind it.