Welcome to the DC CAP AI Program
Enter your full name to begin or resume your onboarding progress. Your progress syncs across devices.
Please enter your full name to continue.
For managers and unit leads: what this investment produces, and how to bring your team along.
Funded by a $600K grant from the KPMG AI Impact Initiative. Designed for replication.
4-tier data classification, acceptable use policy, incident response protocol, and security configuration. Built once, applies to every AI tool the organization adopts.
Ten leaders across six units validated the 4D framework (Delegation, Description, Discernment, Diligence) with measurable pre/post growth. Every DC CAP staff member now trains on the same model.
24+ custom Cowork Skills encoding institutional knowledge into repeatable AI-assisted processes. Student outreach, partner communications, data analysis, and governance QA.
KPI framework with SCALE/PAUSE/PIVOT decision thresholds, weekly pulse data, and pre/post fluency assessment. Evidence base for board reporting and funder accountability.
Claude Enterprise licenses, Innovation Hub staff capacity for program design and technical architecture, interactive training platform development, custom skill engineering, and a dedicated analytics pipeline for measurement.
The governance framework, the 4D competency model (developed by Anthropic), the phased adoption sequence, the facilitation-to-champion handoff model, and the KPI structure. These are the hardest parts to build from scratch. DC CAP built them, tested them, and is documenting the results.
Most enterprise AI rollouts stall for three reasons: no governance structure, no measurement discipline, and no plan for scaling beyond early adopters. This program started with all three.
Anyone ready to engage at this depth belongs here, whether or not you manage people. Most AI rollouts stall in the same place: the tool ships, the training stays optional, and people wait to see what happens. Where someone leans in, adoption becomes real. These six plays are the part the site can't do for you.
No team to lead yet? You still set conditions. Model it on your own work, show a draft that flopped beside the fix, and tell your manager you want to help the next group learn. Read the six plays for the shape of what comes next.
Your team takes its cue from you. Confirm every person has a working seat before Week 0. Say the privacy line out loud in a team meeting: their chats are private, you can't see them, and you won't ask. Show a first draft that fell flat beside the fix. A manager who only shows polished wins teaches the team to hide the misses. And name the weekly challenge as about 45 minutes of work they already owe, run with Claude.
The license is the cheap part. What you get back depends on how your team uses it. Coach away from the two habits that waste it: dumping a whole drive into one prompt, and running every task on the heaviest model. Frame the return in your unit's own terms: an hour back on drafts and data pulls is an hour returned to scholars. Each week you'll see how many of your people used Claude on real work. That is the number that matters.
Two governance calls are yours alone. Project membership: everyone you add inherits the whole knowledge base, so before you add a person or a file, ask whether it clears for everyone already in the room. Connectors: a read-only connection is safe to hand out. A write-enabled one can change a Monday board or a shared file, so you approve which of those your team turns on, and you route new ones through the Innovation Hub.
Not every useful thing should become a maintained build. When a workflow is frequent, high-stakes, or shared across the team, it earns a real Project or Skill with an owner and a data-tier ceiling. When it's a one-off, let people improvise and move on. Watch for the orphaned build: a Project nobody maintains that quietly drifts out of date. If your team ships it, someone on your team owns it.
Each week you'll see how your unit is using Claude. Look for the quiet non-starter: the person who hasn't logged a real task two weeks running. That's a coaching conversation, and it's yours to have. Healthy looks like most of the team trying Claude on real work, with a few starting to build. A team still clustered at "never" by Week 3 is your signal to ask what's in the way.
Preston can't carry this alone. Each unit grows at least one person who can teach the next group: the colleague who took to it fast and likes helping others. Hand them a Friday segment, let them bring the build they shipped, and fade to the back of the room. By the fall, your unit runs its own learning without waiting on the Innovation Hub.
Every Week 0 tour stop also carries a leader note; this section is the playbook behind them. The weekly data you'll read, and the thresholds behind it, live in the KPI Framework. Engagement data never enters performance evaluation.