The Software Factory in the AI Era — Part 2
DRIVE, Five Pillars to Measure Your Organization's Health
In Part 1, we made the diagnosis: AI has created unprecedented forward pressure on our engineering organizations, and we lack the systemic backpressure to master it. AWS, Stripe, and Google showed us the way with their operational excellence reviews.
The practical question remains: what should we measure, and how do we turn those measurements into action? Enter the DRIVE framework.
Why a New Framework?
Unlike existing productivity frameworks (DX Core 4, SPACE, DevEx), which measure individual developer effectiveness, DRIVE measures organizational effectiveness — the right atomic unit for the AI era, where PRs are written, reviewed, and deployed without human intervention. Making one developer 10% more productive is pointless if the organization around them can't absorb, validate, and operate what's being produced.
DRIVE doesn't replace DORA: it incorporates its key signals and adds the governance layer that turns metrics into action.
The Five Pillars
Each pillar answers a leadership question:
D — Delivery: Are we shipping fast, and is it sustainable? Key metrics: deploy frequency, lead time (p95), on-call pager volume. That last one is crucial: excessive on-call burden is a "pull the stop cord" signal, because burnout is irreversible.
R — Reliability: Are we delivering on our promises to customers? Functional SLOs as binary pass/fail, Sev0/Sev1 incidents. AI makes it easy to write tests that give a false sense of confidence — hence the grounding in real customer experience.
I — Initiatives: Are our org-wide engineering investments making progress? Migrations, platform adoption, AI governance... These efforts, perpetually deprioritized in favor of feature work, finally get leadership-level visibility.
V — Vigilance: Are we actively defending our systems? Open critical CVEs, non-compliant assets, and above all orphaned assets — those ownerless repositories that are nobody's problem, and therefore everybody's problem.
E — Efficiency: Are we allocating resources to the right problems? Cloud spend vs. budget, AI token costs (an exploding line item), and the percentage of capacity spent on innovation versus keep-the-lights-on work.
DRIVE at a glance: five pillars, five leadership questions, and the metrics that answer them.
Metrics Aren't Enough: The OpEx Review
Here's the essential point: DRIVE is not yet another dashboard. The metrics only have value when coupled with the Operational Excellence (OpEx) review — that weekly or biweekly, unmissable meeting where the data is interrogated by humans and turned into concrete reallocations of time, people, and money.
A few principles that make the difference:
An aggregated DRIVE report: a table where each domain or team displays its red/yellow/green status across all five pillars, with drill-down capability (VP → directors → managers → services).
A dedicated facilitator who digs into anomalies and asks the uncomfortable questions: "All of your SLOs are green, but this incident report says customers experienced a 15-minute outage. How do these line up?"
A focus on anomalies only: if everything is green, move on. The fastest way to kill an OpEx review is to drown people in data.
A blameless culture, open to everyone — this isn't leadership in an ivory tower, it's the whole organization getting better together.
This is also the best defense against Goodhart's Law ("when a measure becomes a target, it ceases to be a good measure"): only repeated, critical human scrutiny keeps metrics from losing their meaning.
The aggregated report makes the anomaly visible in three seconds. The rest of the meeting is spent understanding it, not finding it.
Where to Start?
No need to deploy everything at once. The recommended approach is progressive:
First, pick the two pillars that matter most to your organization — ideally two that counterbalance each other, like Delivery and Reliability. Start with the metrics you already have, even imperfect ones. Launch the OpEx review to build the organizational muscle, then gradually expand to the other pillars. And automate data collection as fast as you can: the day report preparation lands on managers' shoulders, the review becomes an expensive status update exercise... and dies.
A four-step adoption: two opposing pillars, the data you already have, the review, then expansion.
Drive Fast, With Real Brakes
"Slow down and be careful with AI" is the wrong answer. Slowing down doesn't fix anything: it simply sends back into the shadows the bottlenecks that AI has just revealed. The right answer is to build the systems and discipline that let you drive at full speed — and go a little faster with each lap.
AI has given every 50-person team the output of a 100-person team. But the systems around them weren't designed to absorb that multiplier. The organizations that succeed in the coming years won't be the ones producing the most code: they'll be the ones that learned to improve systematically, on the strength of their systems rather than their heroes.
Drive fast.
Article inspired by the white paper "DRIVE: Operational Excellence for the AI Software Factory" by Ganesh Datta, co-founder and CTO of Cortex.