The challenge

The leadership team had mastered assisted work — Copilot inside documents they were editing themselves. The next step was delegation: handing a defined piece of work to an AI agent and reviewing a finished artefact. For a company that moves money and controls physical infrastructure, that step cannot be taken casually — it needs classification discipline, model economics, and a data boundary set before anything runs.

The approach

  • One day in five blocks: the delegation shift, the machinery, executive judgement, depth on tenders with Fable 5, and landing it in each directorate.
  • Chat versus Cowork as a management discipline: an answer you act on versus a finished artefact you brief and review — with the control point moved from the question to the brief and the review gate.
  • Model routing on economics: Sonnet 5 for daily work, Opus 5 as the default, Fable 5 by exception — justified by decision value, never by how difficult the task feels.
  • The Kahraba control panel: Manual permission mode always, one folder per task, standing instructions carrying the audit clause, and projects per domain — tenders, monthly close, board pack.
  • Skills in four kinds — Doer, Formatter, Checker, Rule — mapped for seven directorates from Executive to Tenders, then built live in a 12-minute exercise; plugins to package them per department.
  • Scheduled tasks demonstrated live: a Monday 07:00 CEO energy brief covering regulation, EGP movement, fuel pricing and competitor announcements, every claim cited.
  • The deep dive: a bid/no-bid tender analysis working section by section through Instructions to Bidders, Bid Data Sheet, evaluation criteria and conditions of contract — every deciding clause cited.

The result

  • Each leader left able to classify any incoming ask as chat work or Cowork work, and to route it to the right model tier on cost and decision value.
  • A stated data boundary — which Kahraba data may enter which tool and connector class — agreed in the room.
  • Directorate-specific skills specified for seven functions, one built live by each participant and edited before saving.
  • Each executive nominated one process from their own directorate for delegated execution — the take-home commitment.

The second engagement — and why it matters more

A client that comes back is the real rating. After 65 executives went through the Copilot programme, Kahraba’s CEO and senior management booked a second, harder day: not using AI inside documents, but delegating whole pieces of work to it — and supervising the result the way an executive supervises a direct report.

From assistance to delegation

The organising idea of the day: chat returns an answer you act on; Cowork returns a finished artefact. Your role shifts from asking questions to writing briefs and holding a review gate. The failure mode shifts too — from an unhelpful answer to confident work in the wrong direction. The session drilled the discipline that manages that risk: literal sources, an audit clause in the standing instructions, “stop and ask” conditions, and Manual permission mode — always, for a company that moves money and controls physical infrastructure.

Model routing as a budget decision

The room learned to read the model table the way a CFO reads a vendor deck: Sonnet 5 for routine volume, Opus 5 as the working default, Fable 5 by exception — for long reasoning chains, large contradictory document sets, and decisions worth a thousand times the token cost. Vendor-reported benchmarks were read next to independent ones, deliberately.

Built for their directorates, not in the abstract

A skills matrix mapped four kinds of skill — Doer, Formatter, Checker, Rule — across seven Kahraba directorates: Executive, Finance, Projects, Operations, Information Systems, Warehouses and Tenders. Each participant then built the one for their own directorate live, edited its SKILL.md before saving — because a skill you did not edit is a skill you do not own — and saw how a plugin packages it for a whole department.

The demo the CEO remembers

Two, in fact: a scheduled Monday-morning energy brief that researches the Egyptian electricity sector, currency movement and fuel pricing before anyone is awake, every claim cited; and a bid/no-bid analysis of a real tender pack, argued clause by clause. Both are the difference between talking about agentic AI and watching it do a week’s work in minutes — with the verification discipline to trust the result.

Talk to Ahmed

Every Menova programme is scoped to the client. Book a short call — or send the team size, seniority and objective — and you get a written proposal with agenda, duration and price within two working days.