MSAI Lab · by 2nth · South Africa, globally relevant

Microsoft AI, done in the right order

Copilot, agents and the estate underneath them — plain-language briefings, a live demo that ends the debate, and open tooling you can run against your own tenant today.

Estate A · what most tenants look like
Monthly Report Final.docx
Monthly Report Final_v2.docx
Monthly Report FINAL-USE-THIS.docx
Client tracker - my copy.xlsx
Board - Confidential (shared: everyone)
Estate B · governed
2026-03 Operations Report Final
Client Register List · typed
Permissions named groups
Versions SharePoint's job
Board pack restricted
Same tenant. Same Copilot. Same question. Different answers.
01

The one idea that matters

AI assistants answer from your company's filing. Copilot reads what your business already has — documents, spreadsheets, email — and answers from it. That's the whole product.

If the filing is five versions of the same report and folders shared to everyone since 2023, you'll pay per person, per month, for an assistant that confidently summarises the wrong version. The assistant isn't broken. The filing is.

The IT tidy-up you've deferred for years and the AI programme are the same project. Nobody sells you that sentence, because the fix isn't a licence — it's work.

AI, Explained for the CFO No jargon: one idea, four numbers, one spending rule. The briefing to read before signing anything. 20-question readiness self-assessment Score yourself honestly — it tells you what to fix first, and what AI can safely do today. Copilot's honest limits Five limits to hear before the seats land — and why none of them are reasons not to buy.
02

The maturity scale

Five levels. Each gates what AI can safely do in your business. The most common failure is level-skipping: buying Level-4 seats while operating at Level 1.

1Ad hocPersonal drives, email attachments, spreadsheets as systems of record — where most organisations start
2OrganisedGoverned filing and permissions — Copilot becomes safe to switch on
3StructuredReal databases behind the documents — AI answers from data, not guesswork
4AugmentedAI embedded in daily work, adoption measured — the 12-month target
5AgenticGoverned agents doing multi-step work — humans approving, not assembling
03

Microsoft AI & tech, explained

Plain-language explainers from the 2nth knowledge tree — written from real deployments, with a South African lens on residency and licensing that travels well anywhere.

LiveCopilot in South Africa Six Copilot surfaces compared — a decision framework for which to deploy, where, for whom. LiveCopilot Studio — agents are software Version-controlled, tested, deployed: ALM patterns that treat agents as real components, not toys. LiveDataverse The governed data platform under Power Platform and Copilot Studio — and when to keep it lean. LiveMAI & Humanist Superintelligence Microsoft's frontier-model strategy: seven models, four pillars, and a when-to-adopt framework. LiveAzure OpenAI from Workers Calling frontier models in-country from Cloudflare Workers — the headless residency pattern. SoonAzure AI Agent Service The full-code orchestration layer: threads, tools, file search — where pro-code agents land. SoonM365 Copilot — SA geo deep-dive Tenant verification and data-leakage surfaces for the South African geography. Browse the full Microsoft branch → All Microsoft explainers on know.2nth.ai, with related branches: Cloudflare, ERP/CRM, legal & residency.
04

See it, don't debate it

The two-estates demo puts the same AI against two versions of the same business — one with messy filing, one governed. Same question, radically different answers. Ninety seconds.

It ends the "just buy the licences" conversation in every room it runs in — and it's open source, so you can build it inside your own tenant in about ten minutes — the full script, prompts and gotchas are on the demo page.

The kit includes the fictional sample business, seeding scripts (Graph-only or PnP), and the build log from running it against our own tenant — gotchas included. Scripts live on GitHub.