# Changing the work, not just the tools

By Tommy Magee

Type: Writing
Canonical: https://tommymagee.com/library/when-everyone-starts-building-agents

A firsthand account of enterprise AI implementation: workflow redesign, agents managed by subject matter experts, and the ownership that lets teams build.

The most useful AI work I’ve led started with a question about the business: how should the work change now that AI can do some of it? At Rivian, my team and I helped business teams rethink how they plan, coordinate, and make decisions across capital programs and corporate real estate.

I connected the process, people, and data framework with implementation: shaping the agent platform, mentoring teams building their own solutions, and establishing ownership for what they built. The work was shared. Keeping those pieces connected was a central part of my job.

## Start with the information

We mapped how information moved between people and systems: what came in, what needed to come out, and where it went next. Doing that with the people responsible for the work helped us question the steps themselves.

We used three lenses:

- **Process:** Which steps still need to exist?
- **People:** What can this team change, and what crosses into someone else’s responsibilities?
- **Data:** What information does the job need, and can people and agents use it?

Some processes could go away. Some work belonged with an outside provider. Other workflows needed a change in responsibilities before better software would help. Being able to automate a task wasn’t enough reason to keep doing it.

## Put control close to the work

We built a platform of custom agents that subject matter experts could tune and manage themselves. They understood the job. Depending on developers for every adjustment would have limited what they could do with it.

The work reached across cost and portfolio management, site selection, leasing, cash modeling, forecasting, and workplace operations. We connected systems through custom APIs and Model Context Protocol (MCP), with Slack as a place people could work across them. Making the underlying data readable by AI and accessible to people building tools was a major engineering priority.

## Give teams room to build

I created AI councils and a champions program to find people who combined subject matter expertise with an aptitude for AI. My team and I mentored them, developed building kits, and created a tool registry to make ownership visible.

Contract review support was one of the workflows champions worked on. Other tools supported finance and accounting, meeting records, and dashboards that reduced coordination between groups. The people who knew the work could build with help from my team.

A personal tool, a shared team solution, and an enterprise initiative need different support. As others began relying on what teams built, we had to decide where to invest, where effort was being duplicated, and who would look after it. That became part of enablement, not something to sort out later.

## What changed, and what is still ahead

We eliminated unnecessary processes, simplified the technology stack, and reduced tooling and licensing spend. Redesigned workflows and automation also freed capacity and helped teams absorb work without new hires. Lower spending is a saving; released time gives people room to take on other work.

Our planning points to more capacity for a growing capital program. The next test is how this way of working holds up as that work arrives.

## What I would carry into another business

Finance, contracts, forecasting, and coordination are familiar problems well beyond construction and real estate. The useful starting point is the same: follow one piece of work, understand who needs what, and decide which responsibilities should change.

I evaluate the whole job, including the checking, correction, and support an agent adds. I’ve turned those questions into an [evaluation worksheet](/resources/agent-workflow-field-test) you can adapt.

The frameworks and change management are the part I’m proudest of. They gave us a way to improve shared systems while helping teams solve problems for themselves.

Topics: Enterprise AI, Organizational change, Capital programs, AI implementation
