Before AI can act, it needs the right context

Adapted from an article by David Brudenell, Decidr Co-CEO and Executive Chairman of Decidr AI Industries.

Recently, I was at a conference just outside San Francisco, listening to OpenAI’s CFO talk about practical adoption and the gap between what AI now makes possible and how companies are actually using it.

OpenAI has publicly described its 2026 focus as “practical adoption,” with the priority being to close the gap between what AI makes possible and how people, companies and countries use it day to day.

The word that kept coming up wasn’t models. It wasn’t compute. It wasn’t tokens.

It was context.

The same theme came up again in Australia. At the AFR AI Summit 2026, Telstra CEO Vicki Brady said AI’s next frontier is capturing a company’s human “secret sauce”: the know-how, context and judgement that sit inside its people. Different stage, same problem: capturing the knowledge that makes AI useful inside a real company.

How do you give AI enough structured understanding of a company that it can do useful work? How do you move from impressive output to reliable execution? How do you stop asking a model to guess what your best people already know?

I sat there thinking, that’s our entire thesis.

Decidr’s been building toward this for two years. That’s why we acquired Sugarwork, to find how work actually gets done, beneath the org chart and outside the system of record.

And it’s why we’ve now acquired Rumi, to capture the knowledge in the conversations where work happens every day.

Sugarwork gives us the map. Rumi keeps it alive.

The AI orchestration layer starts with context

AI orchestration is the system that sits between frontier models and real work. It gives AI structure, routing, memory, governance and auditability.

Without that layer, most enterprise AI is still prompt-and-pray.

You send a question to a powerful model. It produces something fluent. The answer might be useful. It might be wrong in a way that looks useful. It might miss the commercial context, the compliance history, the regional nuance, the customer politics or the unwritten rule everyone on your team knows.

More tokens don’t solve that.

A senior underwriter evaluating a complex claim draws on pattern recognition built over years. A procurement lead knows which suppliers actually deliver on time and which ones pad their quotes because she’s watched them do both. A regional sales director reads a room in Jakarta differently from a room in Melbourne, and that difference can decide the deal.

That knowledge rarely sits cleanly in a database. Much of it’s never been written down. This isn’t a data problem, it’s a knowledge problem, where the decisions, context and expertise needed for AI sit in people’s heads rather than systems.

And it’s the thing that makes the company work.

Sugarwork finds the hidden map

Sugarwork was the first step.

We acquired Sugarwork because it performs the archaeological dig. It finds how work actually gets done inside a company, including the version hidden behind the org chart.

Sugarwork captures decisions, context and expertise, then maps how work actually works, including judgement calls and workarounds.

Sugarwork interviews people, maps workflows and surfaces the tribal knowledge that never made it into a document. It captures the handoffs, workarounds, exceptions and judgement calls experienced people carry in their heads.

That map is valuable before AI touches it.

Most companies don’t know how they really work. They know how they think they work, which is a very different thing. A Sugarwork deep dive gives leaders visibility into the risk and value of each task in a workflow, and where AI or automation could actually help.

For many teams, that’s the first honest answer they’ve had to a basic question.

What do we actually do all day?

This is also why Decidr treats workflow discovery as the foundation for orchestration. Before you can build reliable AI execution, you need to understand the process architecture underneath it. The question is whether companies understand their processes well enough to automate them. More AI doesn’t fix weak process.

Rumi makes the map live

Rumi is the second step.

Rumi.ai is a San Diego-based AI meeting intelligence platform. Rumi’s products, X-Ray and Meeting Memory, capture conversations across meetings, in-person catchups, Slack and CRMs, then surface the decisions, commitments and reasoning patterns inside them.

That matters because companies don’t stay still.

An onboarding process that worked six months ago has already been patched by three people who never updated the documentation. A workaround someone invented for a broken CRM integration last quarter has become the default process. A pricing exception made in one market starts shaping sales behaviour in another.

The workflow map changes.

Sugarwork captures the deep structure. Rumi captures the ongoing signal.

Together, they turn workflow discovery from a point-in-time exercise into a living map.

That’s the missing piece in most enterprise AI programs. They treat knowledge as something you upload once. In reality, knowledge is being created, revised and contradicted all day, in meetings, messages, side conversations and decisions that never make it into a formal system.

Rumi gives businesses a way to capture that signal continuously.

From workflow discovery to live context

Sugarwork and Rumi give a business the context it needs to orchestrate work in Decidr’s schema.

Sugarwork gives Decidr the starting point: a structured view of how work actually happens within the customer’s business. It surfaces the workflows, handoffs, exceptions and judgement calls that usually sit outside the system of record.

Rumi keeps that view current. It captures the decisions, commitments and context being created every day in meetings, Slack threads and working conversations.

Decidr builds on both. It takes the workflow map from Sugarwork and the live signal from Rumi, then turns them into structured context its agentic apps can use.

That means work doesn’t start from a blank prompt. It starts from an understanding of the goals, rules, history and decisions that matter.

Sugarwork shows Decidr how a customer’s work happens. Rumi shows Decidr how that knowledge changes. Decidr uses both to make AI execution more consistent, traceable and owned.

Knowledge security is becoming a board issue

There’s a security dimension here that companies are only starting to understand.

Every prompt your team sends to a third-party model reveals something about how your company thinks. Even where enterprise agreements restrict training on customer data, the pattern of usage still says a lot.

What gets asked. How problems get decomposed. Which decisions need escalation. Which exceptions recur. That reasoning structure took years to build.

Rumi and Sugarwork help capture it inside your own boundary. That’s the strategic shift.

You stop leaking context into tools you don’t control. You start building an intelligence asset you own. That’s the heart of knowledge security, and it’s why sovereign AI models will matter more as companies move from experimentation to execution.

A sovereign AI model isn’t just about where computation happens. It’s about who owns the operational memory being created.

The first principle is still the same

The fix isn’t complicated. It’s just hard to do well.

Understand your company before you automate it. Capture institutional knowledge inside your own boundary. Structure that knowledge so machines can reason against it.

That’s why Rumi matters. Sugarwork maps how work happens. Rumi keeps that map current. DecidrOS turns it into execution that’s structured, traceable and owned.

There’s an old saying about eating an elephant.

One bite at a time.

Enterprise AI is the same. You don’t transform a company by pointing a model at a mess and hoping it understands you. You do it one workflow at a time, with the context captured, the tasks structured and the execution owned.

That’s the work now.

That’s why context matters.

If your team is already using AI but still struggling with inconsistent execution, undocumented knowledge or unclear workflows, DecidrOS and Sugarwork can help you find where orchestration should start.