You’re Not Automating a Process. You’re Automating the Brokenness.

Author: David Brain

Three of the biggest names in AI just spent billions on a surprisingly human solution.

Not better models.

People. Engineers whose job is to sit inside a business and make the technology actually work.

This month, Anthropic and Blackstone launched Ode, an implementation venture built specifically for mid-sized companies. OpenAI raised roughly $4 billion for its own Deployment Company in May. AWS followed with a $1 billion forward-deployed engineering organization in June.

The message is hard to miss. The model is no longer the hardest part. Getting it to work inside a real business is.

But there’s a problem even the best embedded engineer can’t solve alone: most companies can’t fully explain how their own work gets done.

Real work is full of exceptions, workarounds, handoffs, and knowledge that lives in someone’s head rather than a process document. As the magentIQ co-founder Ian Barkin puts it:

Ian Barkin: “Automating your mess for less just gets you a faster, broken process.”

That doesn’t mean every legacy process is badly designed. Ian is careful about this: “Not to say that these processes are broken. They’re just optimized to the environment in which they have to survive.”

Which is the point. The brokenness isn’t bad design. It’s the distance between the constraints a process was built for and the ones you’re operating under now.

An agent can’t see that distance. It doesn’t know which workaround is outdated, which approval is essential, or which exception could cost you a customer. It executes what it’s given, faster.

How to Rank Your AI Ideas in a Week

Appetite for AI. No shared view of where to start. That’s the gap the AI Vision Sprint is built to close. Two hours spread across a week: a short pre-read, two calls, and a survey in between, so you hear from the people doing the work, not just the loudest voices in the room.

You leave with a ranked priority list and an AI-readiness score across five pillars.

Which means you’ll actually know where to start.

See how the Sprint works: Explore the AI Vision Workshop Sprint

Process debt compounds at machine speed

HFS Research puts unresolved process debt at $7.7 trillion in enterprise value, missed revenue and cost reductions businesses never capture.

For a COO, that’s duplicated work, unclear ownership, and exceptions bouncing between teams. For a CFO, it’s the same problem with a number attached: implementation costs that keep growing, gains that never reach the P&L, and employees spending their “time saved” checking and correcting machine output.

The mistake is measuring the automated task in isolation.

A reconciliation runs in seconds, but what happens when the data doesn’t match? A customer response generates instantly, but who decides whether it’s commercially sensible? A report arrives faster, but can leadership trust the assumptions behind it?

The task is automated. The outcome isn’t.

Companies are already reacting. Retool’s 2026 research found 35% of enterprise teams have replaced at least one SaaS product with something custom-built.

That instinct makes sense. Off-the-shelf software is built around an average version of the work, and your business isn’t average to you.

But building custom around an unmapped workflow just makes the existing process more permanent and harder to see. You preserve the unnecessary approval. You reproduce the workaround. You encode the exception nobody questioned.

Which brings us back to the engineers. Every implementation announcement reads the same:

• “We’ll embed an engineer in your business.” An embedded engineer isn’t the same as a mapped process.
• “We’ll move fast.” Fast just means you find out what’s broken faster.
• “We’ll de-risk your AI rollout.” De-risk it how? Scaling a process you haven’t examined doesn’t reduce the risk. It multiplies it.

Load-bearing or scar tissue?

So, the first question isn’t where to add an agent. It’s simpler, and much harder to answer: which parts of this process would you defend if someone asked why they exist?

Ask it of every step. The approval that exists because of one incident in 2015. The reconciliation that runs because two systems never talked to each other. The exception one person handles from memory and has never written down.

Some of those steps are load-bearing. Some are scar tissue.

Until you know which is which, you’re not choosing what to automate. You’re choosing what to make permanent.

That’s also how you decide what stays with a person, not as a fallback for when the automation fails, but because judgment, context, and accountability were never really about speed.

How do we actually do this?

Four things to get right before you scale anything.

1. Pick something people actually complain about. A pilot can tick every logical box, saves 0.6%, integrates cleanly, passes procurement, and still go nowhere. The ones that build momentum solve a pain people recognize and talk about afterwards.
2. Write down why you started. Then don’t move it. Put the reason on a whiteboard and circle it: do not erase. Most pilots get judged months later against a goal nobody wrote down, and get called failures for missing a target they were never set.
3. Keep the task small enough to validate. “Here’s a problem, give me an answer” isn’t a workflow. If you can’t see how it reached the answer, you can’t validate it. Give AI a discrete task, then flow the output back into your own business rules, with validation points and confidence scores.
4. Don’t mistake 70% for nearly done. Getting most of the way there in an hour doesn’t mean the rest takes another hour. The last 30% is where the exceptions live, and it’s the part that’s hard to get after.

MagentIQ co-founders Ian Barkin and David Brain go deeper on all of this in Episode 6 of The magentIQ Show, listen on Apple Podcasts, Spotify, or iHeartRadio.

The hype phase is over

The hype phase told businesses to deploy more AI. The execution phase belongs to the ones who know what their work actually costs them before they scale it.

Before you automate the process, make sure you’re not automating the brokenness.

Rapidly discover which parts of your process are worth automating, fixing, or removing in a FREE Vision Workshop Sprint.