Specialization: what everyone needs to know and what depends on the job
A common core for everyone, plus selective depth where your judgment is hardest.
10.1The common core
What every AI-era engineer needs
Competent adoption starts with deep use of existing tools and with improving the development environment — not with building an agent runtime, training models or mastering all ML infrastructure.
- Formulate tasks
- Provide context
- Operate tools
- Choose verifications
- Interpret results
- Control access
- Learn from failures
- Keep the fundamentals of your systems
10.2Three scopes of work
Which one is yours?
Each scope adds priority knowledge on top of the core — and has its own practical evidence of mastery.
10.3Selective depth
Breadth to follow a delivery, depth where judgment is hardest
AI lets you broaden your area of work, but professional commitment should stay proportional to your ability to verify consequences.
Data
Integrity, queries, schema evolution
Frontend
Interaction, accessibility, perceived performance
Platform
Reliability and isolation
Systems
Resources and concurrency
Key takeaways
- 1
The common core is about using existing tools deeply and improving the environment — not building ML infrastructure.
- 2
Know your scope: building with AI, building platforms for agents, or building AI-powered product behavior.
- 3
Combine breadth with selective depth where your judgment is hardest, and bring in expertise where it is missing.
Check your understanding
Question 1 of 2
When should a product engineer learn RAG or fine-tuning?
Read the full section of the original reportThe complete, unabridged text — with every source and nuance.
10.1. The common core
Every engineer who uses AI professionally should be able to formulate tasks, provide context, operate tools, choose verifications, interpret results, control access, and learn from failures. They also need to maintain the fundamentals relevant to the systems they are responsible for.
This does not require building an agent runtime, training models, or mastering all of the machine learning infrastructure. Competent adoption starts with deep use of existing tools and with improving the development environment.
10.2. Three scopes of work
| Professional scope | Priority additional knowledge | Practical evidence of mastery |
|---|---|---|
| Engineer who develops software with AI | Context, delegation, verification, permissions, and delivery flow | Delivers a complete change with appropriate autonomy and reliable evidence. |
| Platform engineer for agents | Environments, integrations, identity, isolation, shared evaluation, and observability | Multiple teams carry out tasks with less setup and intervention. |
| Product engineer whose behavior depends on AI | Behavior evaluation, information retrieval, data quality, latency, cost, and failures of the system in use | Demonstrates success on user tasks and tracks regressions in production. |
For the third scope, techniques such as RAG, model fine-tuning, and specialized orchestration should be learned when they solve an observed problem. Their value depends on requirements for freshness, accuracy, access, data distribution, and cost. An implementation of these techniques also needs its own evaluation.
10.3. Selective depth
A sustainable education combines enough breadth to follow a delivery end to end with depth in the areas where the engineer needs to exercise hard judgment. The data specialist deepens integrity, queries, and schema evolution; the frontend specialist, interaction, accessibility, and perceived performance; the platform specialist, reliability and isolation; the systems specialist, resources and concurrency.
AI makes it possible to broaden one's area of work, but professional commitment should remain proportional to the ability to verify consequences. When a task crosses an area of impact that no one masters sufficiently, the competent response includes bringing in the appropriate expertise and recording the decision.