Editor’s note: The following is a guest post from Michael Dennis, partner, enterprise architecture at Envorso.
In the 1985 horror film Silver Bullet, Marty Coslaw and his Uncle Red melt a silver crucifix and a medallion into a single bullet. One shot, and the werewolf is dead.
It’s a great scene. But also a work of fiction — in the real world, there is no silver bullet. Yet executives who were in their youth when that movie came out keep trying to buy one.
For organizations, the beasts are real enough: Lagging efficiency, sluggish competitiveness, quality that costs too much. But AI is not the bullet. Rewiring how work gets done is the hard part, and the only part that pays off.
We know this because it already happened once.
Agile started as a revolt against processes that smothered the work, and it delivered: the Standish Group’s Chaos research found Agile projects succeeded more than three times as often as waterfall ones, which is exactly why the approach became popular.
But then it got hijacked by frameworks, certifications, and coaches who had never shipped anything. Ceremonies meant to help became status theatre. Twenty-plus years later, the phrase “we’re Agile” tells you nothing about whether a team can actually deliver.
AI is now retracing those steps. But the blast radius is much bigger this time, because it reaches every job built on knowledge work. However, there are ways to ensure the approach goes beyond just changing the vocabulary and actually reaping results.
Redesign the work, don’t decorate it
Most AI failures come from bolting a model onto a process built for humans doing every step by hand. That just makes a broken workflow slightly faster.
The teams getting real leverage tear the workflow down to its purpose and rebuild it around what the tool can now do, removing handoffs, review layers and steps that only existed because the old way was done at a human’s pace.
Just 39% of organizations report any EBIT impact from AI, according to McKinsey, most of them under 5%. MIT also found most enterprise generative-AI pilots delivered no measurable return on the P&L, a failure it attributes to integration issues and a learning gap — or organizations bolting AI onto workflows that were never redesigned to use it.
McKinsey found the other half of the same coin. High performers who do capture impact are nearly three times more likely to have fundamentally redesigned their workflows, not layering AI onto old ones.
Put the spotlight on outcomes
When adoption gets measured in seats, logins and percent of staff using AI, organizations are focusing on vanity metrics. These numbers tell boards that people are busy, but not that anything got shipped, met the demand, increased revenue or shipped with quality.
Metr, the AI research organization, ran a randomized trial in early 2025 with experienced open-source developers and found that tasks took about 19% longer when AI was allowed, even though those same developers believed they had worked roughly 20% faster.
When Metr ran the experiment again later that year the estimates flipped toward speedup: about 18% faster for the 10 developers who returned from the original study, and about 4% faster for 47 newly recruited ones.
Neither result was statistically clean, both confidence intervals straddled zero, and there was a deeper problem with the data. A growing share of developers declined to take part rather than spend half their time working without AI, and 30% to 50% of those who did admitted withholding exactly the tasks they most wanted AI for. A selection that all runs one way, which is why Metr treats its estimate as a lower bound. The real gain is probably larger; Metr just can’t say how much larger.
This means even an organization whose entire job is measuring AI capability struggled to clearly determine the size of the technology’s effect.
The true lesson is that businesses should track the things that carry their own evidence: cycle time, cost per unit of work, output per person, quality and revenue. If you can’t point to a number that moved based on the work itself, that means you have a pilot rather than a real result.
The biggest gains show up when one person or a small team can own a piece of work end-to-end, because AI collapses the coordination overhead that used to require a whole department. That only happens if teams have the authority to change how they operate. This will not happen if every decision routes back through a steering committee.
Looking back at the Agile adoption wave, the people who got value from the technology changed how they worked. The rest just changed what they said in standup meetings. AI is identical. Incentives shouldn’t reward looking busy, protecting headcount or following old processes.
Instead, leaders should reward the teams that redesign, that ship more with less and kill their own busywork.
There was never a silver bullet. There was only ever the work. So skip the theatre. Don’t ask how many people are “using AI.” Ask what shipped, measure the real results, and reward the teams that changed how the work gets done.
