Every enterprise I talk to right now wants the same thing: agents. Autonomous workflows. Something that routes the lead, cleans the record, updates the forecast, and does it while the team sleeps. The pitch is everywhere, and the technology is finally good enough to deliver on a lot of it.
So why do most of these deployments quietly stall three months in?
Here's the uncomfortable answer we keep running into: the agent usually works fine. The thing underneath it doesn't.
When a company buys an AI agent, they're picturing the demo, clean data in, clean action out. What actually happens is the agent gets dropped onto a CRM that's been duct-taped together for six years, an ERP that speaks a different language, and a marketing platform that updates contact records on its own schedule for reasons nobody fully remembers. The agent does exactly what it's told. It just gets garbage.
We call the stuff holding those systems together "human glue" the rep who copies a deal from one tool to another, the ops person who runs a Tuesday-morning script to fix duplicates, the analyst reconciling two dashboards that should match but never do. Most companies don't see this work because it's invisible until it breaks. It's also the exact work an agent inherits the second you switch it on.
Automate a broken process and you don't get a fixed process. You get a broken process running faster, with less oversight, at scale.
I'll be concrete. A freight brokerage we worked with had lead routing that took 18 minutes on a good day. They wanted an agent to handle it. The problem wasn't the routing logic — it was that two of their three lead channels wrote to fields the CRM didn't map cleanly, so about a fifth of leads showed up with the wrong source, wrong owner, or no company match. An agent routing on that would've sent good leads to the wrong place, confidently, faster than any human. We started with the mapping, not the agent. Once the systems agreed on what a lead record was, routing dropped under 90 seconds and then the agent had something solid to stand on.
That order is the whole thing. Architecture first. Agent second. Always.
Smart teams skip this step, not because they're careless, but because nobody gets budget approved by promising six weeks of reconciling field definitions. You get budget by promising autonomous revenue workflows. So the sequence reverses, the agent goes in first, looks great in a pilot with clean data, then meets the real data and starts making confident mistakes. By month three, trust is gone and the team has quietly gone back to doing it by hand.
The technology didn't fail. The foundation was never poured.
On July 9, we're running a live session, Beyond the Bot: Overcoming the RevOps Infrastructure Gaps, walking through exactly where these break and what the layer underneath should look like.
Sign in to leave a comment.