The Metric That Stopped Measuring You

The Metric That Stopped Measuring You

Status
Draft
Author
Created
Jul 16, 2026
Tags
Agentic AI
Human Judgment
Consulting Ethics
Text
Dashboards are greener than ever. Here's why that stopped being a compliment.
A managing director I know ran her midyear review last week and watched every dashboard turn green at once. Tickets closed faster. Campaigns shipped in half the time. Code merged before lunch. She should have felt triumphant. Instead she felt uneasy, and she couldn't say why until she looked closer: the targets that used to separate her best people from her average ones had all quietly become achievable by anyone who knew how to write a decent prompt.
That's the tension sitting underneath most performance conversations right now, even if nobody's named it out loud. For decades, output was a reasonable proxy for skill. If you closed more tickets, wrote more code, produced more content, you were probably better at your job than the person next to you who did less. Agentic AI has severed that link. Karl Chan, CEO of Laserfiche, described this exact moment after his company's 2026 midyear cycle: teams shattered every execution-based metric they had, and the shattering wasn't a signal of excellence — it was a signal that the metric had stopped measuring people at all. If an employee can hit 100% of an output target by feeding a tool a prompt, that target isn't testing human capability anymore. It's testing whether they have API access.
This should worry anyone who runs a team or sells advisory services for a living, because most of our machinery — performance reviews, engagement pricing, promotion criteria — is still built on the old assumption. Consulting has been especially exposed here. A huge share of billable work has always been "produce the deliverable": the deck, the model, the memo, the analysis. Agents now produce a passable first draft of most of that in minutes. Firms who keep charging and promoting on volume are, whether they realize it or not, pricing and rewarding a capability that's rapidly becoming free.
The instinct is to panic about this, or to look for a new number to replace the old one — some cleverer output metric that AI can't game as easily. That instinct is wrong, and it's worth being precise about why. The problem was never that the old KPIs were poorly calibrated. The problem is that output, as a category, is no longer where the human contribution lives. You cannot fix a category error by picking a better metric inside the same category. You have to move to a different one entirely.
The reframe that actually holds up is this: AI collapses the distance between "started" and "produced." It does not collapse the distance between "produced" and "correct," and it does nothing at all for the distance between "correct" and "right for this client, this context, this moment." Those two gaps — produced to correct, correct to right — are now almost the entire job. Everything upstream of them is commodity. This lines up with what researchers studying human-AI collaboration keep finding: decision quality, not throughput, is what actually differentiates strong performers once a baseline of AI-assisted output is available to everyone. It also lines up with something practitioners are starting to say plainly — that the real measure of AI use isn't frequency, it's judgment.
So what does this mean concretely, rather than as a nice sentence in a strategy memo? It means the questions worth asking about a piece of work have changed. Not "how fast did this get done" but "what did the human catch that the agent missed." Not "how many drafts shipped" but "how many of those drafts would have embarrassed us if a client had seen them unedited." Not "did the agent hit the target" but "who is accountable for the target being the right one to hit in the first place" — because agents can now execute against a bad objective just as tirelessly as a good one, and someone still has to own the judgment call about which objective was worth executing at all.
For engagements specifically, this argues for a structural change, not a cosmetic one. Price and staff around critique, not production. Build in an explicit review layer where a senior person's entire job is to interrogate the AI-generated baseline — where it's generic, where it's wrong for this specific client, where it optimized for looking finished over being useful. That layer is not overhead anymore. It is the product. The deliverable was always meant to be a vehicle for judgment; for a while, producing the vehicle was hard enough that we forgot judgment was the point.
The dashboards will keep turning green. That was never in question — it was never really a measure of us to begin with. The harder, more useful question is what we do with the time that used to disappear into producing the thing, now that producing the thing takes an afternoon instead of a week. Whatever fills that space is the actual job now. Everything before it was just the cost of admission.