"Holy crap. The end of me": What Eric Schmidt's AI moment means for product teams
Eric Schmidt watched an AI generate a complete program on its own and said, "Holy crap. The end of me." He's been coding for 55 years. If he's calling time on parts of the craft, product teams should pay attention.
His broader point: AI isn't just writing code. It's coming for the operational work that quietly eats your budget-billing, accounting, product design steps, delivery, and inventory. That's where the near-term value sits.
The signal: AI is moving from helper to owner of tasks
At leading labs, Schmidt says AI is already writing 10-20% of code. That share will keep climbing. He also expects systems that learn and plan with less human guidance to arrive within a few years. That doesn't remove humans, but it will reassign them.
Translation for product: treat AI as a teammate that takes full tickets, not just autocomplete. Your roadmap should reflect this shift.
Where the value shifts inside companies
- Finance ops: Invoice generation, matching, fraud checks, and reconciliation.
- Support and success: Auto-triage, resolution drafts, and deflection via grounded assistants.
- Supply chain: Demand forecasts, reorder proposals, and exception handling.
- Product and design: Requirement translation, spec drafts, and component stubs.
- Engineering: Test generation, refactors, code reviews, migration scripts.
None of this is flashy. It saves millions because it cuts cycle time and error rates across the line items that never make your keynote.
90-day plan for product leaders
- Week 1-2: Pick three processes with high volume and clear rules (e.g., support macros, invoice checks, test case creation). Define success metrics (time saved, cost per task, accuracy).
- Week 3-6: Ship narrow pilots with a human-in-the-loop. Goal: 30-60% automation with measurable quality gates.
- Week 7-10: Add evaluations and rollback paths. Track precision/recall, latency, override rate, and cost per task.
- Week 11-12: Graduate one pilot to production with access controls, audit trails, and incident playbooks.
Operating model upgrades
- PRDs for AI features: Define the source of truth, allowed data, guardrails, and failure modes. Include an escalation policy.
- Human oversight by design: Set thresholds where a person must review or approve.
- Risk framework: Map use cases against an external standard like the NIST AI Risk Management Framework.
- Change management: Don't drop automation on teams. Co-create workflows and share win rates weekly.
Technical notes that save time
- Start small: Retrieval + prompts + structured outputs (JSON) will cover more than you think.
- Data first: Clean event logs, unify IDs, and version prompts. If the inputs are noisy, quality will drift.
- Evals: Build a lightweight eval set per use case. Run it on every change like unit tests.
- Observability: Log prompts, responses, costs, and user overrides. You can't improve what you can't see.
Team skills to invest in
- Product: Problem selection, prompt specs, risk thinking, and metrics.
- Engineering: Tooling for evals, data contracts, and API orchestration.
- Design: Conversation flows, affordances for uncertainty, and clear handoffs to humans.
- Ops: QA protocols, sampling, and incident response.
What to pilot next week
- Support assistant that drafts replies from your knowledge base, with agent notes and confidence scores.
- Automated test generation for critical services; gate merges on passing rates.
- Spec-to-stub: turn user stories into API skeletons and acceptance tests.
- Invoice matching agent that flags exceptions with reasoning and suggested actions.
- Weekly product insights: auto-summarize user feedback and create prioritized issue lists.
On timelines and oversight
Schmidt sees general AI-level capability arriving in roughly three to five years, with systems learning more on their own. Whether you agree or not, the safe move is optionality: build the pipeline, keep humans involved, and measure relentlessly.
His warning stands: someone has to say, "we went too far." As you scale automation, make that threshold explicit and review it quarterly.
What this means for your roadmap
- Shift scope: From features to throughput. Your product wins if it cuts cycle time across the business.
- Budget for data and evals: Treat them like core infra.
- Measure real outcomes: Time saved, defect rates, customer effort, and margin-not demo sizzle.
- Keep humans central: Oversight is a feature, not an afterthought.
If you want structured upskilling for your team, see our AI certification for coding or browse tools that assist with generative code.
Your membership also unlocks:
