ETA · Enterprise Technology Accelerator
Measure how work really moves. Then automate it.
ETA (Enterprise Technology Accelerator) is a measurement-first operating layer for work that crosses teams. It records how cross-functional work moves through the tools you already run, shows what it costs, then automates the waste you agree, with a person approving every change. It is in design, and there are no ETA client results yet.
How it works
From scattered tools to one measured workflow.
Agents arrived in every function and the silos stayed. ETA starts by measuring how long a cross-functional workflow takes, before it automates anything.
01
Connect
ETA reads the tools you already run. Nothing to rip out or replace.
02
Map
The meaning a workflow needs (cases, steps, who acted) is mapped onto your tools, in place, not copied out of them.
03
Measure
A work ledger records each step as the work runs. You see how long a workflow takes, who touches it and what it costs.
04
Approve
Agents propose changes and a person commits them. Autonomy is earned step by step, on evidence.
Five design rules
Five rules that don't change.
ETA is designed around these rules. Each one trades a bigger promise for a safer one.
It runs in your cloud
The work runs in your tenancy. Anitech holds metadata only: no payloads, prompts or business data.
Mapped, not copied
ETA maps meaning onto your systems. It doesn't build a second copy of your data.
Reading is separated from acting
What an agent may read and what it may change are separate permissions, each logged.
The harness proposes. A person commits.
This caps the savings story: the gains are in cycle time, fewer errors and released capacity, not headcount.
Everything is recorded
Every step an agent takes and every decision a person makes goes into the work ledger.
Instrument forward
If ETA runs the work, the record of it is a by-product. We don't rebuild history from old system logs.
Governed autonomy
Autonomy is earned per step, not granted per agent.
Promotion needs evidence and a named owner. Demotion is automatic when errors breach a threshold, when an upstream process or model changes, or when the override rate falls to zero, because that usually means people stopped checking.
Shadow
The agent proposes and nothing reaches a system. We measure how often it agrees with people.
Supervised
A named person approves each action. We watch the override rate, not the approval rate.
Delegated
Low-consequence steps run unattended, with tiers set per class of step.
Autonomous
The agent acts within an approved boundary. People supervise by exception and by sampling.
Build, buy, ride
We build the method. We ride on your platform.
ETA is a layer on your estate. We own what makes the measurement credible and leave the rest to what you already run or can buy.
The part that is ours
The delivery method, workflow instrumentation, connectors for the long tail of systems and the economics of what each workflow costs.
The part that is yours
Your platform's agents, identity, governance and audit, used deliberately rather than rebuilt.
The part not worth owning
Commodity integration. We don't build our own integration platform.
Phased delivery
Each phase has a gate. The next one starts when the evidence says so.
Measure
Instrument a handful of workflows and cost them in people and elapsed time. Nothing writes back to your systems.
Automate agreed waste
Rules and agents handle the waste you've chosen. People approve what is written.
Operating model
Agents act within boundaries you set, and the economics stay instrumented.
Gates, not dates
A phase ends on an exit test agreed in advance. We publish no timetable and no availability date.
Hosting
Hosted in your cloud. A second option we would only offer with a change to rule one.
In your cloud
ETA's work runs inside your own tenancy and security boundary. Our control plane holds metadata only.
Hosted by Anitech (considered, not our default)
This would break rule one, so we would only offer it with that change stated in the contract and in the next version of the white paper. Today ETA runs in your cloud.
What we don't claim yet
The limits, stated up front.
No ETA results yet
There is no ETA pilot, so this page publishes no result for it. The first thing we'd give you is a measurement, not a promise.
Matching one case across systems
Following a single case across systems without hand-mapping the join keys is an open problem for us. Whether we can solve it decides how ETA is delivered.
It won't suit every workflow
A workflow needs a ground-truth signal, steps that are reversible or bounded, and inputs that can be retrieved. Tasks of equal difficulty can land on opposite sides of what agents do well.
A programme before a product
Today ETA is delivered as a method with our people alongside you. It is not software you install on day one.
Read more
The full argument, and the thinking behind it.
A governed operating layer for work that crosses functions
The diagnosis, the work ledger, the five rules, governed autonomy, the architecture and the honest limits, by Jack Dinneen and Andrew Nolan.
Agents arrived. The silos stayed.
Why measurement has to come before automation.
An override rate of zero is an alarm
How autonomy is earned, and why a falling override rate can mean people stopped checking.
Built on ANI
ETA is Artificial Neural Intelligence, in product form.
Your systems, data and teams behaving like one nervous system, with your outcome at the centre.
Start with a measurement.
Tell us which tools your teams run today and which workflow nobody can time. We will tell you whether ETA fits.