Adé IjidakinroAI Enablement · Program Management · Learning & Workflow Design
New York · Open to remote opportunities
I’m a program manager with experience in learning technology, product design, and enterprise IT. At Ehoro Village, I develop AI-supported workflows, support team adoption, and use learner data to guide improvements. My work combines practical implementation, clear documentation, and human review.
01 Lanes
Pick the lane you're hiring for. The proof is the same work, seen from a different side.
I help teams adopt AI workflows they keep using: the workflow itself, the training, the documentation and the review step.
For AI enablement, AI adoption, technical education and customer education roles.
I scope AI projects, set decision rules and review gates, and keep delivery documented so anyone can follow what was decided and why.
For program manager, enablement program manager and technical program roles.
I design learning and workflows people can follow: needs analysis, action mapping, facilitator guides and job aids, and the process behind them.
For learning experience design, curriculum, and workflow or process design roles.
I coordinate AI agents across Claude, ChatGPT, Gemini and DeepSeek, and I vibe-code the prototypes: each agent owns a lane, handoffs are written down, and a usage meter stops runs before they hit the limit.
For AI operations, internal AI tooling and agent workflow roles.
An octopus keeps about two-thirds of its neurons in its arms. Each arm works on its own; the head sets the intent. That's how I run AI agents.One owner of intent. Many arms that think. A written rule for when they disagree.
02 System
In plain terms: one model makes decisions, one plans, several build and test, and I approve. It runs on the same habits I bring to teams: decision logs, review gates, written handoffs and a budget.
03 Platforms
Field notes from running them side by side on the same work. Not benchmarks: what each one was good for, and the rule I wrote afterward.
Best at careful judgment and clean merges, so it held the rulings and the only merge key. It also ruled from memory once, and the ruling had to be withdrawn.
Rule nowCite the source before ruling. One model merges; the others propose.
A tireless planner and runner. Its status reports understated problems three times, and a five-agent run burned 98% of a 5-hour window in one night.
Rule nowEvidence over status. One session per lane, and every run stops on the meter.
A strong second opinion. It ran the smoke tests on finished pieces, and handled research and longer documents in long solo runs.
Rule nowDon't let the builder test its own work. Give testing to a different model.
Fast and cheap for breadth: market scans, source checks and research briefs. It ran a full shift until it hit the meter line.
Rule nowSpend expensive models on judgment and cheap ones on breadth. Research arrives as a brief; another agent commits it.
04 Experiments
Every rule in the system came from a failure. Four of them, with the numbers.
Five agents shared one board and coordinated through claims and heartbeats: 259 commits in 7 hours, 146 of them in a single hour.
Lean lanes: one session per lane, one commit per finished ticket, bookkeeping once at handoff. A script reads the real usage meter and stops the run before the limit.
The planning agent's status reports understated problems three separate times.
Reports lead with failures, attach evidence and prove absence. I check the repo, not the summary.
A design ruling was made from memory instead of the written source, and had to be withdrawn later.
A clerk gate asks “already written or already ruled?” before anything reaches the decision-maker, and rulings cite sources.
Agents kept producing drafts and prep work while the main queue waited on five answers only I could give.
The queue asks the human first: a ten-minute question list unblocked the held work. A ping budget caps interruptions at three a week.
05 Work
My work at Ehoro Village first, then independent builds that show the technical side. Open any row.
My first multi-agent build, and where the Overlord / Underlord method came from. I set the direction, made the creator decisions and relayed every packet. Claude Opus ruled, ChatGPT planned, and Claude Sonnet, Codex, Gemini and DeepSeek built, tested and researched. Work moved through eight phase gates, and every finding had to be fixed, kept on purpose, sourced or deferred with a reason. Unreleased.
Built for faculty. An instructor pastes an assignment and rubric; three agents attempt it the way students commonly use AI, and an advisor agent writes a vulnerability report with specific revisions. Status: functional prototype.
I designed, facilitated and evaluated a 75-minute workshop built on one framework: Frame → Generate → Audit → Log. In a two-person pilot, both participants caught more planted flaws with the audit checklist than without it. Evaluated at Kirkpatrick levels 1–3; level 4 wasn't measurable at pilot size.
Cleans mixed-method feedback, turns comments into signals and ranks churn risk by segment, with the reasons shown. It scrubs personal details and labels results as directional when there are fewer than 200 responses. Status: MVP.
Waveform slicing, a step sequencer and a live looper in the browser. No install, no signup.
06 Experience
Program management came first. AI made it faster; it didn't replace it.
Ehoro Village
AI-supported course production, team adoption and documentation, hands-on workshops for 20+ internal champions, and a Python/Streamlit review tool that cut manual triage 40–60%. I use learner data to guide improvements.
University of Michigan · BRAID
Research with 100+ stakeholders; a redesign that lifted new sign-ups 40% in four weeks. Case study →
Procter & Gamble
Coordinated 11 regional project managers on a $20M+ global initiative; scaled a framework to 10+ teams; CEO Award.
University of Michigan · Multi-Ethnic Student Affairs
Nearly four years of heritage months, seminars and workshops, including anti-racism workshops for classes of 40+.
07 About
“I design like sushi: few high-quality ingredients, applied in a way that makes sense.”
I came to AI through learning design and program management, and I still work that way. AI speeds up the scripting, the code and the busywork; people make the calls. The method on this page grew out of a playbook I wrote across 24+ AI build sessions.
Fun fact: I make beats, and I'm allergic to carrots.
08 Contact
I’m open to AI enablement, program management, and workflow improvement roles in New York or remote.