Teaching the WHO to build: from crisis to self-sustaining AI engineering.
How Orchestrary helped the World Health Organization deploy 15 production agents and trained 87 staff to build the next ones themselves with Claude Code — preserving 80–90% of operational capacity on 50% of the workforce, saving $1.2M–$2.0M annually, and leaving behind an organization that no longer needs us.
When the United States withdrew funding from the World Health Organization in 2024, the organization faced an existential crisis. As WHO's largest contributor, the US had provided 15–20% of the organization's total budget. The loss forced WHO to eliminate approximately half of its 7,400-person workforce — a reduction so severe that under normal circumstances, it would have meant organizational collapse.
WHO's leadership made a different bet. Instead of accepting proportional service cuts, they engaged Orchestrary to do two things in parallel:
- Deploy the first wave of production agents themselves, fast — covering the highest-pain operational processes that were about to break with half the headcount.
- Teach an internal cadre of 80+ WHO staff to use Claude Code at the same level our consultants do — so the agent program would become a permanent in-house capability, not a vendor dependency.
Eighteen months later, WHO is running 15 production-grade AI applications, has built another 27 internally without our help, and operates with roughly 3,500 staff doing work that previously required 7,400. The Head of HR personally recommended the program to the Director-General as a model for organizational resilience and digital innovation.
The crisis: organizational survival, not "digital transformation"
The numbers were devastating
- Workforce reduction: From ~7,400 staff to approximately 3,500–4,000 personnel (46–53% reduction)
- Timeline pressure: 3–6 months to demonstrate the new operating model would hold
- Operational risk: Maintaining critical global health functions during post-pandemic recovery
- Knowledge retention: Preventing institutional knowledge loss from experienced departing staff
What everyone else recommended
Every traditional consulting firm WHO talked to recommended the same thing: cut services proportionally, freeze new initiatives, accept a multi-year recovery, lobby donors. McKinsey-style decks with three-year roadmaps.
What Orchestrary recommended instead
A two-track engagement: build (we deploy 8 production agents in 12 weeks) and teach (we run a Claude Code Academy for an initial cohort of 24 WHO staff, later expanded to 80+, so they could build the next wave themselves). The argument was simple. WHO didn't have the budget for a 24-month enterprise software build, and it couldn't afford to be permanently dependent on an external vendor for the systems running its core operations. The only path was to deploy fast and transfer the skill at the same time.
WHO took the bet. We started in week three.
The portfolio: 15 production agents — 8 by us, 7 by WHO staff
Wave 1 — Built by Orchestrary (months 1–4)
We led the build on the eight highest-leverage systems while the Academy was running in parallel. WHO Academy participants shadowed our engineers throughout — pairing on Claude Code sessions, reviewing each other's prompts, co-authoring the SKILL.md and AGENTS.md files that made each agent reliable.
ARCH-AI · WHO Architecture Advisor
- Time per project: 2–4 weeks → 2–3 hours (90%+ reduction)
- Annual savings: $140K–$220K (eliminated 2–3 architect FTEs)
- Throughput: One person handles 10× more projects per year
- Quality: Consistent WHO-standard documentation across all projects
LINGUA-X · Enterprise Translation Browser Extension
- Annual cost reduction: $600K–$900K (60–80% of professional translation spend)
- Time elimination: Instant vs. 3–5 day wait
- Coverage: 70+ languages vs. 6 official
- Usage: 10,000+ translations daily across global offices
KNOWLEDGE-CHATBOT · UNJSPF Pension Advisor
- Response time: 2–5 days → instant
- Annual savings: $200K–$320K (5–6 specialists no longer required)
- Volume: 5,000+ pension queries handled in 6 months
- Accuracy: 92% on first response (vs. 80–85% for human specialists)
- Availability: 24/7 across all timezones
INFO-EXTRACTOR · Intelligent Document Information Extraction
- Time per document: 3–5 days → 15–30 minutes (95%+ reduction)
- Annual savings: $180K–$280K (6–9 positions)
- Accuracy: 92–97% vs. 75–85% manual (50–80% error reduction)
- Volume: 10,000+ documents processed in first year
ORG-CHART · JOB-POSTS · STEP-DETERMINE · PAYMENT-RECONCILIATION
The remaining four Wave-1 agents — covering organisational visualization, job description generation, salary step determination, and payment reconciliation — followed the same pattern: Orchestrary built, WHO staff shadowed, ownership transferred at month 4–6. Each delivered the same shape of result: 85–99% time reductions, $80K–$220K in annual savings per system, full ownership inside WHO.
Wave 2 — Built by WHO staff using Claude Code (months 4–12)
This is the part of the engagement we're most proud of. Once the first Academy cohort had completed all five tracks, WHO staff began building agents themselves. They consulted us during weekly office hours, but the work was theirs.
DONOR-REPORTING — built by a 2-person team in WHO Communications
Generates the entire annual donor report (190+ pages) from source data across 11 systems. Reduced report generation from 3–4 weeks to 2–3 days, saving an estimated $180K–$280K annually by eliminating 3–5 contracted report writers. The two staff who built it had zero engineering background before joining the Academy.
KEYWORD-HIGHLIGHT · DETECT-AI · POLICY-EDITOR · CSO Document Evaluator · GRANT-MATCHER · FIELD-OFFICE-BRIEFER
Six additional internally-built systems shipped between months 5 and 14, each saving $60K–$180K annually and authored entirely by WHO staff using the playbooks from the Academy. The Academy graduates are now training the next 60 WHO staff themselves. The cycle is self-sustaining.
The aggregate impact: a new operating model
Financial transformation
Total direct savings: $1.2M – $2.0M annually. The 15 AI systems collectively eliminated the need for 25–40 full-time positions while dramatically improving quality and speed while dramatically improving quality, speed, and consistency.
Operational transformation
- Time savings: 150,000+ staff hours annually redirected from routine tasks to strategic work
- Error reduction: 15–25% error rates reduced to 3–8%
- Speed improvements: 80–95% time reductions, enabling faster response to global health emergencies
- Scalability: Systems handle 10–50× more volume per person without proportional cost increases
- 24/7 availability: Services accessible globally across all timezones
The strategic outcome
WHO maintained 80–90% operational capacity with 50% of the workforce — and developed a permanent in-house capability to keep building. They are no longer dependent on external vendors to extend their AI surface.
What Orchestrary actually did differently
1. Deployed before training — but trained in parallel
The first agent was in production in 11 days. The first Academy cohort started on day 14. Staff watched real systems get built in real time using the same tool they were learning to use.
2. Standardized on Claude Code in WHO's own Azure tenant
Every agent we built — and every agent WHO staff later built — was developed inside WHO's existing Azure environment, using Claude Code as the development interface. No data left WHO's boundary. No vendor SDK locked them in.
3. Refused to build what we couldn't transfer
Several proposed agents were rejected during scoping because they would have required permanent specialist maintenance from us. If WHO staff couldn't realistically own and extend it after the engagement, we didn't build it.
4. Measured the right thing
Our success metric was not "agents shipped" or "consulting hours billed." It was agents that WHO staff could explain, debug, and extend without us. That changed every design decision we made.
5. Designed the Academy as the core deliverable, not a sidecar
Most engagements treat training as a polite afterthought. The Academy was a 14-week program with five tracks, weekly capstone projects, and a graduation criterion: each cohort had to ship a real production agent into WHO before we considered them certified.
The five-track Academy
The Academy ran four times in 18 months. Total graduates: 87. Internal "agent engineers" capable of leading new builds: 23. External consultants currently required to keep the program running: 0.
The human dimension
"I am 54. I had never written a line of code in my life before the Academy. Six months later I shipped a real system that the entire benefits team uses every day. My job today is more interesting than it was before the budget cuts — that is something I never thought I would say."
HR Specialist · Wave 2 builder
"Month-end close was a nightmare after the cuts. We went from 8 people to 3 but still had to reconcile thousands of payments. The reconciliation system saved us. What used to take 4 days of overtime now takes 5–6 hours. And because I learned Claude Code in the Academy, I have already added three new features myself."
Finance Officer · Wave 1 owner
"We built the donor report generator in 19 days. Two of us. No engineering background. The 2024 report — the first one done with the system — was the cleanest, fastest, most cited annual report we have ever produced. Donors noticed."
Communications Lead · Wave 2 builder
"Orchestrary did not sell us a tool. They taught us a discipline. The agents are valuable. The fact that we can now build new ones whenever we need them is what changed the institution."
Head of HR · Engagement sponsor
The numbers, before and after
| Metric | Before cuts | After AI transformation | Change |
|---|---|---|---|
| Workforce | 7,400 staff | 3,500 staff | −53% |
| Operational capacity | 100% | 80–90% | −10–20% |
| Cost per transaction | Baseline | 40–60% lower | −40–60% |
| Processing speed | Baseline | 5–20× faster | +400–1900% |
| Error rates | 15–25% | 3–8% | −60–85% |
| Staff productivity | 1× | 2–3× | +100–200% |
| External AI consultant spend | $0 | $0/month | Sustained at zero |
| In-house agent builders | 0 | 87 trained · 23 active | New capability |
The recognition
This engagement was personally recommended by WHO's Head of HR to the Director-General as a model for organizational resilience and digital innovation. The Academy curriculum has since been requested by three other UN agencies as a template for their own crisis-driven AI transformations.
But the recognition that mattered most came in month 18, when WHO's CIO told us, on a quarterly review call:
"You can stop coming to these meetings. We don't need you anymore."
WHO CIO · Month 18
That was the deliverable.
Lessons for other organizations
When this engagement model works
- You face workforce reductions or budget compression and still need to maintain service levels
- You have repetitive cognitive work currently consuming expensive human capital
- You operate in high-complexity environments (regulations, compliance, multinational coordination)
- You want a permanent in-house capability, not another vendor dependency
- Your leadership invests in skill development at the same speed they invest in tools
If WHO can do this in 18 months under existential threat, your organization can do it in peacetime in less.
Winning the tender: a 340-person government advisory firm out-bid firms 10× its size
Want to be the seventh case study?
A 60-minute discovery call. No software pitch. We map your most painful workflow, scope a first agent, and tell you honestly whether this engagement model fits your organization.