From Prompting to Orchestration: Decidr AI and the Case for Governing AI in Business

Decidr AI Executive Director David Brudenell returns to Vantage Point to explain why businesses are moving from prompts to governed workflows, and why he believes AI is entering what he calls its sovereign era.

The World Has Changed Since Last Time

David Brudenell, Decidr’s Executive Director, last appeared on Vantage Point more than a year ago, and Simon Hinsley opened by noting how much has moved since. There was no Claude Code to speak of, no OpenClaw and no “token maxing” then. David says Decidr has grown up as a business over the same period. Its partnership strategy is starting to compound, its beta program is now live, and the company had a good last quarter.

What Decidr Does

In plain terms, David says Decidr is an AI infrastructure company that orchestrates the new intelligence. It sits on top of OpenAI, Anthropic and the open source models, and puts them into workflows for businesses.

He contrasts that with how most people use AI today, which is on-demand prompting. The human types into Claude, and a good prompt is hard to write, which is why social media is full of tips on how to prompt. Decidr’s view is that this is a very human way to use AI, like training an employee. Its approach is to break work down to the workflow level, where each workflow is a series of tasks.

When you actually go from task to task, what we would call the prompt becomes very simple. For example, get a first name.

— David Brudenell

At each step, Decidr decides what goes there. Sometimes it is AI, sometimes a human and sometimes just a calculation, like summing a pipeline value. David expects the world to move from prompting to tasks very soon, which narrows what AI is being asked to do.

He describes three layers of what people are now talking about: prompt engineering, context engineering, which is the context of your business and how you do things, and harness engineering, which is governing it all together. Decidr sits in the harness and context layers, which he also calls orchestration or AI coordination.

Skating to Where the Puck Is Going

Simon asked how Decidr has adapted in a market that moves so fast. David quotes Wayne Gretzky about skating to where the puck is going, and says co-founder Paul Chan has been doing exactly that. The strategy, he says, has not changed from the start. The thesis is that AI is just a tool and not a dazzling intelligence.

What has changed is how Decidr talks about it. Two years ago the world was discussing agents, so the company embodied AI in a human form. Then the conversation moved to always-on agentic activity, and now to OpenClaw and Claude Code, where people realise they can create almost anything, but should they?

Just ’cause you can doesn’t mean that you should do things.

— David Brudenell

As these tools connect to private systems like email and internal software, and as token costs rise, people are starting to think they could do some jobs more cheaply with a person. David says that is the start of the sovereign era.

The Sovereign Era of AI

David calls sovereign a catch-all word, but he defines it as governance, accountability and visibility. It means being able to say what the AI did, how it thought, and whether it would do it the same way if asked again.

Simon raised the compounding problem he sees when using Claude, where a mistake in the second of ten tasks gets worse by the tenth. David says that is the problem most people are running into.

As you stack what we would call probability agents together, call them super guessers, you start to compound errors.

— David Brudenell

Because Decidr designs the workflow up front and uses AI for a single simple task at a time, results come back accurate every time, he says. That combats what he sees as the two big problems with AI today: accuracy and cost.

The Boring Layer That Makes It Work

David uses a series of analogies to explain where orchestration fits. The first invention in any new technology is dazzling, he says, but a coordination layer always emerges after it. The internet does not work in your home without the router in the cupboard, which you set up once and never think about again. International payments depend on the SWIFT system, which is boring but critical to scaling.

He believes orchestration is the next coordination layer, and Simon summed it up as AI growing up. David adds that by historical standards AI is moving logarithmically fast, noting that OpenClaw did not exist at the start of the year.

The Beta Program: Mercha and the Agentic Economy

David says Decidr has four revenue lines all built on the same underlying technology, and that they lead to its third strategic pillar, compounding. The big ambition is a new agentic economy, in the way the digital economy changed bricks-and-mortar stores. Businesses in it will look different, so the question is how to bring them onto it.

The beta program deliberately selects businesses across sectors, such as accounting, finance, manufacturing, marketing and sales, so Decidr can refine its workflows to the last mile. Once an agentic business is switched on, it starts sending proposals to customers, which creates new connections and leads. The program is now on its third cohort, and the first is shipping live products.

A recent case study was Mercha, a drop-ship merchandising company that makes swag for businesses. It has about 35,000 different products, so building a proposal manually means deciding which products suit the client, in what quantity, with what discount and shipping. Decidr built the workflow and what David calls a company brain, which pulls purchasing, customer, vendor and transaction data into a single source record.

It’s gone from their capped ability to send 2 to 3 proposals out a day to effectively unlimited.

— David Brudenell

The system might notice a construction client is going to conferences and suggest branded pens and notebooks, plus high-vis vests for the builders and smarter vests for corporate staff. The human still handles the last mile. David says the data alone surprised Mercha’s leadership, because a vendor placed in a lower tier turned out to be connected to a large share of revenue.

His summary of the approach is a familiar saying: how do you eat an elephant? One bite at a time. How do you transform a business into an AI company? One task, or one workflow, at a time.

SugarWork: Capturing What Is in People’s Heads

SugarWork is the recently acquired knowledge-capture platform. David says it captures how you actually do things in a business. Manuals describe how to use a system, but about 80% of a business’s knowledge sits in people’s heads, in the shortcuts and experience that jump between systems. SugarWork extracts that into documented workflows.

It serves two purposes. For enterprises, it protects knowledge. David argues that when employees prompt OpenAI and Anthropic, those companies learn how the business does things, and points to those providers moving across industries. SugarWork helps businesses capture that knowledge themselves, and Decidr is now investing in helping businesses build their own large language models.

If you create your own large language model, it’s between 20 and 30 times cheaper than Anthropic, but it’s also yours, and it’s always on.

— David Brudenell

The second purpose is strategic for Decidr. Tacit knowledge works as a configuration system for your own agentic business, and the settings are made through language: conversations with AI or colleagues, meeting transcripts and the like. David’s vision is not one model for everything but the right model for each task. You do not need a Ferrari, as he puts it, to get a first name.

On how it works in practice, David says SugarWork is about 90% software and 10% services. The team goes into an enterprise, usually starting with a business unit or workflow, and sets up the platform in the customer’s own environment with single sign-on and security. AI interviewers and facilitated guides drawn from research then run what he calls the archaeological dig to capture the knowledge. After that, always-on telemetry keeps learning in the background, with the business controlling what is captured. David jokes that it passes the creep-zone test.

The Moats: Embeddedness and the Agentic Graph

David describes two layers of moat, and both come down to embeddedness. The first is the partner strategy. Decidr builds AI agents or apps for large software companies, which distribute them to their customers, so Decidr becomes an intrinsic workflow in their software. More than 15 partners have signed and five have been commercialised, including CareerOne in classifieds and Ebev in hospitality.

The second is the agentic graph. David compares infrastructure companies laying fibre or building 5G towers as quickly as possible. His best analogy is Google, which had to index all the websites to give good search results, because you cannot index just a couple. Decidr wants to pre-board every registered business in a country onto its graph, grabbing public information to build each one’s agentic business.

He imagines receiving a proposal from a Mercha customer for a set of vests that are cheaper than you expected. You click through to Mercha’s agentic business, switch on your own, and are offered an incentive to look at prospects for your own proposals. It grows like a search graph or social graph, with friends bringing in more friends, and David says it has positive churn.

Where the Growth Comes From

Simon noted that Decidr does not provide forecasts, and asked what drives value. David points to five consecutive quarters of growth since commercialisation. More than half of the growth comes from the partnership channel, with the rest from beta customers and one large customer, the Australian Institute of Management. The business is still Australian only.

He sees compounding opportunities in three places. One is expansion into Singapore, the broader Asia-Pacific region and the US through the SugarWork acquisition. The second is SugarWork itself. David celebrates the first Australian sale, to a very large aged care company with about 3,000 employees, made 45 days after the acquisition settled by a newly trained salesperson. The third is the partners, which typically take 12 to 18 months to reach their net penetration rate.

Two Years Out: Nobody Sees Decidr

Asked what Decidr looks like in two years if everything goes right, David starts with the world. AI will be deeply embedded in many workflows in enterprise businesses, and companies will run differently, probably with fewer meetings and dashboards.

If we’re successful, nobody will see Decidr. Decidr will sit underneath everything.

— David Brudenell

He returns to the router analogy. There is one in every Australian home, which is why Cisco is very happy. The magic number, he says, is 10% penetration in any country. Even two or three per cent of US businesses would be a huge number of companies, he adds, noting that Salesforce has only about 150,000 customers. The goal is for Decidr to be ubiquitous, helping people make use of a remarkable technology.

Simon closed by saying he learns a lot each time they talk, and promised not to wait another 18 months to have David back, perhaps in tuxedos.