I’m worried that they put co-pilot in Excel

AI-Induced Spreadsheet Failures & Accountability

• Several comments predict a major AI-caused financial blow‑up, possibly bankrupting a public company and damaging AI vendors’ reputations.
• Others argue leadership will treat AI as a scapegoat: take credit when things go well, blame “the AI” when they don’t – but that only works once before boards lose patience.
• Some think systemic actors won’t allow large AI failures and will effectively “bail out” AI errors; others respond that audits and existing controls should catch bad numbers regardless of tool.

The “Brenda” Archetype, Jobs, and Institutional Knowledge

• “Brenda” represents the experienced spreadsheet power user who quietly holds together messy, business‑critical processes. Many say she’s irreplaceable due to tacit institutional knowledge and accountability.
• Others present alternative Brendas: low‑context “spreadsheet people” doing duplicate data collection that should be automated away; they argue we need fewer Brendas and more people who can redesign processes.
• There’s extensive debate about incentives: why would Brenda automate herself out of a job when she isn’t rewarded for long‑lived automation, only for ongoing manual work?

Human vs AI Errors: Determinism, Verification, and Audit

• A major theme is determinism: traditional formulas and scripts are predictable and debuggable; LLMs are non‑deterministic, can hallucinate plausible numbers, and may “fudge” outputs.
• Commenters stress that verification is now the central problem: AI can generate slop faster than humans or tools can check it.
• Many note that human spreadsheets are already full of bugs, but human error modes are familiar and often caught by peers, auditors, or sanity checks; AI error modes are harder to predict and detect.

Current Reality of Copilot and AI in Office Tools

• Multiple participants report that Copilot in the Microsoft stack is mostly useless today: disabled features, weak transcription, poor handling of domain acronyms and names, and unhelpful Excel behavior.
• Some find LLMs useful for learning what’s possible in Excel or for simple code/scripts, but say they fall apart on messy, real‑world workbooks.
• There’s pushback against chat‑box UIs jammed into every workflow; people want AI that integrates naturally and preserves transparency, logs, and change tracking.

Automation, Excel, and AI’s Place

• Excel is described as the de facto programming environment of the economy, often doubling as database, app platform, and glue between incompatible enterprise systems.
• Some argue agents will eventually bypass Excel entirely—querying databases directly and generating reports—automating both Brenda and spreadsheets. Others reply that real companies run on fragile legacy stacks and undocumented “one‑box” scripts; AI cannot simply drop in.
• Overall sentiment: AI in Excel could be helpful under human‑in‑the‑loop use, but over‑trusting it in finance and operations without strong audits, versioning, and cultural skepticism is seen as dangerous.