Telstra's AI control plane is a governance product

First you let a hundred flowers bloom. Then the bill arrives. Then someone like Chris Ellis has to build a plane that can see cost, risk and whether the thing even works. You've probably lived that sequence: a year of experiments, a surprise invoice and a sudden appetite for governance. Ellis is Head of corporate software engineering at Telstra. Telstra built an AI control plane with real-time visibility of performance, risk and cost. Not a chatbot. A dashboard for an estate that has already escaped the lab.

Ellis said the challenge is no longer managing individual deployments. It is governing an increasingly enterprise-wide AI ecosystem. At an SAP event he said they can no longer put AI out in the wild and let the organisation experiment. They have to productise outcomes and be prescriptive. I have waited years for a telco to say "prescriptive" in public. Most still sell empowerment and then act shocked at the bill.

Ackland's 380 and the lock-in fear

CFO Michael Ackland's half-year is the denominator. Three hundred and eighty use cases. The risk he named: licences, cloud and AI providers eating the benefits. They are streamlining cloud. They do not want to be locked into any one large language model. That is an adult procurement sentence. If your CAIO cannot say it, you are in a vendor's roadmap, not a programme.

The work sits under Connected Future 30 and a workforce re-engineering story. Ellis said change how the organisation operates, not just automate tasks. Telstra is working with SAP and Infosys. He talked about completely rethinking the future of work and throwing the process rulebook out. I will keep that sentence only because it is attached to a control plane and a CFO who can see 380 things at once. Without those, it is a town hall.

If you run a large Australian estate, Telstra is describing your next two years. First you let a hundred flowers bloom. Then the bill arrives. Then someone like Ellis has to build a plane that can see cost, risk and whether the thing even works. Enablement is not optional in that picture. People who cannot operate inside a prescribed pattern will keep putting AI in the wild. The named Australian work is on the site if you want a client-shaped version of sitting in the process: Sportsbet.

Ackland's fear that licences, cloud and model vendors offset the benefit is the CFO version of "we automated the meeting notes and the bill went up." Streamlining cloud and refusing a single-model lock-in are the two controls you can copy this quarter without waiting for Infosys. Multi-model is not a philosophy. It is a commercial hedge. Ellis still has to productise the outcomes or you just have three invoices.

Connected Future 30 is the programme name. Workforce re-engineering is the people name. Do not let them live on different packs. A control plane that cannot see who is allowed to run a use case is just financial operations with extra steps. The SAP event line about being prescriptive is the culture change. Most telcos still reward experiment theatre. Ellis is saying the theatre is over.

I would put 380 on a wall and force each one to declare an owner, a cost line and a kill date. The ones that cannot will be why the control plane exists. Throw-the-rulebook-out only works if a new rulebook ships. Otherwise you have a slogan and 380 surprises.

Ellis working with SAP and Infosys is named, so name them. It is not a secret skunkworks. It is a telco admitting the estate is too big for a team of heroes. The control plane is how the heroes get replaced by a pattern.

What this means for L&D

One control plane, three live metrics (performance, risk, cost). Print 380. Name Ellis and Ackland. Multi-model on purpose. Stop calling experiment a strategy once the CFO can count the use cases. The operators who will live inside the prescribed pattern are the training job. The dashboard will not teach them.

Sources: iTnews, 31 August 2026.