Microsoft Fiscal Year 2026 Second Quarter Earnings Conference Call
Copilot, Azure, OpenAI exposure, agentic computing language and workforce efficiency pressure.
Microsoft Fiscal Year 2026 Second Quarter Earnings Conference Call
FIRST LINE:On the call with me are Satya Nadella, chairman and chief executive officer, Amy Hood, chief financial officer, Alice Jolla, chief accounting officer, and Keith Dolliver, corporate secretary and deputy general counsel.
The Butcher's Version
Microsoft just stood on an investor call and calmly described the factory farm for white-collar labour. Not AI assistance. Not productivity. Not empowerment. A slaughterhouse with governance controls.
The old software model was simple: sell tools to workers. The new model is cleaner and uglier: sell agents to management so they need fewer workers. That is the whole trick.
What Satya Actually Said
When Nadella says agents are the new apps, he is saying the unit of enterprise software is no longer the human user. It is the synthetic worker.
Apps waited for humans. Agents eat workflows. Apps made employees faster. Agents make employees optional. Apps were hammers. Agents are replacement carpenters.
The agent control plane is not a cute governance feature. It is the HR department for machine labour: identity, permissions, monitoring, security, audit trails, compliance, memory and access control.
The Corpse on the Table
The first body on the floor is the junior knowledge worker: analyst, associate, coordinator, paralegal, SOC triage grunt, sales ops person, marketing assistant, junior developer, admin, reporting analyst and support worker.
These were the people who did the messy first-pass work that trained them into senior people. Microsoft's pitch is not framed as why pay them, but that is the mechanical implication once agents can draft, model, cite, format, triage and execute.
The Verification Trap
The human does not disappear immediately. That is the comforting lie people cling to. The human becomes the verifier.
At first: review the AI output. Then: spot-check the AI output. Then: handle exceptions. Then: why do we still need a full-time person for exceptions?
The verifier role is the anaesthetic before the amputation. It lets the organisation pretend there is still a human-centred workflow while the cognitive production has moved into the machine.
15 Million Copilot Seats
Workers hear Copilot seats and think employees get tools. Management hears each employee can absorb more workload, so the company does not have to backfill attrition.
That is the ratchet. A company does not need to fire half the office on day one. It stops hiring, merges roles, lets people leave and does not replace them. One person gets five agents and the corpse is called operational efficiency.
The Real Product Is Dependency
Microsoft is not merely selling AI. It is selling institutional dependency.
Once email, documents, meetings, code, security logs, CRM, ERP, compliance records, identity and agent permissions all run through Microsoft, leaving is no longer a procurement decision. It is organ removal.
Microsoft is not a vendor in the AI transition. It is one of the landlords of the successor system. Tenant companies pay rent in tokens. Workers pay with relevance.
Tokens Per Watt Per Dollar
This is the most brutal metric in the call. The metric is not employees per revenue. It is not productivity per worker. It is tokens per watt per dollar.
Translation: cognition has been reduced to an energy-efficiency problem. Your brain is now competing with a datacentre's thermal envelope.
You need sleep, salary, healthcare, context, motivation, legal protection, management, onboarding and emotional handling. The machine needs electricity, cooling, chips and a model update. Capital does not see a moral dilemma. It sees waste.
Final Verdict
Microsoft's call is not about quarterly performance. It is a map of the kill chain.
First own the compute. Then own the models. Then own the data layer. Then own the agents. Then own the governance. Then sell every corporation the ability to replace its cognitive workforce while calling it transformation.
The worker is not being upgraded. The worker is being abstracted. Once you are abstracted, you are priced. Once you are priced against tokens, watts and dollars, you are already meat in the spreadsheet.
Oracle Read
Extracts
Microsoft frames AI as broad GDP diffusion rather than a narrow product cycle.
The transcript repeatedly joins AI demand, capex, GPUs, CPUs, data-center capacity and cloud supply constraints.
The agent section turns software work into an infrastructure problem: tools, orchestration, context, observability, security, compute and storage.
The labour question is absent from the call, which is exactly the tell: Microsoft is building the substitute production layer and reporting it as cloud growth.