Sweep It Under the Bot

"As companies rush to adopt AI for speed and efficiency, many are discovering that technology also exposes deeper operational weaknesses. Fragmented workflows, unclear ownership and inconsistent communication get scaled up alongside efficiency gains." — Jessica Wong

I often joke that I am my father's favorite son, which is funny because I am both his daughter and his only child. He did not have a son to hand the flashlight to while he worked on projects around the house, so he handed it to me. Same with the lawn mower, the trash cans, the spare tire and whatever else needed doing.

I remember building a bed frame with my dad in my first apartment on a boiling July day. The thing had come in no fewer than 183 pieces, and the "instructions" consisted of a diagram whose primary function was to show you that you were wrong, but not how. I grabbed a screw and a drill and lined them up quickly, impatiently. We were so close to done, and my patience had burned away with the afternoon sun.

My dad stopped me. "That screw isn't straight."

The lean was almost invisible. I could not imagine it mattering. "It's straight enough," I argued.

He shook his head. "If you start with it barely off, you'll end with it 45 degrees from straight."

He was right, as he often is. I think about that lesson more than I expected to as AI spreads across business operations.

Firms are deploying AI for contract analysis, research compression, billing compliance, pitch drafting. The tools are getting measurably better. The discourse is focused on what AI can do. What gets discussed less is what AI reveals, and occasionally entrenches, about the operational conditions it gets deployed into.

Processes that are fragmented (even subtly so) do not usually get fixed by AI, they get scaled.

Unclear ownership, the kind where three people are all technically responsible for something and none of them are actually responsible, does not disappear because an AI system is tracking the workflow. The AI follows the ambiguity. Inconsistent communication patterns, the billing team that operates one way and the LPM team that operates another with a gap between them that nobody formally owns, do not get resolved by adding an AI summarization layer. You get AI-generated summaries of the miscommunication.

Put in a legal ops context and the risk is easy to see: a firm adopts AI-assisted budget monitoring. The tool is good, it flags overage risk early, tracks actuals against estimates, sends alerts when a matter is trending off-plan. If the scoping process upstream of the budget was imprecise to begin with, the monitoring system now generates very accurate alerts about a budget that was never realistic. The precision is not helping. It is documenting the failure more efficiently.

Before we rush to automate a workflow, running it manually long enough to know where the friction is, who owns each step, and what happens when something goes wrong is likely a wise choice. A few questions worth asking before any legal ops AI implementation: if this process failed tomorrow, who would know? How quickly? Who would fix it? What are the handoff points, and are they documented in a way that a new team member could follow without a twenty-minute verbal explanation? Is the measure of success tied to output quality or output volume?

These are not rhetorical. Running through them takes an afternoon. Untangling the process failures they reveal takes considerably longer, but better before they are baked into automated workflows running at scale.

My dad would be insufferable about being right on that one.