4 Levels of AI Practitioner
A competency framework for patent attorneys: from quality checks to running your own agents.
Article last updated on 2026-04-29
Core message: AI competence is also a question of work organisation. The upper levels separate themselves from entry-level practice through systematic integration into daily work, not through better prompt tricks.
Why a level model?
AI adoption in every firm is a patchwork: some team members experiment with ChatGPT, others stick with classical search. A level model makes the progression visible and gives every participant a direction: where am I, and what would the next step worth taking look like?
The model is tailored to patent practice and deliberately kept lean. The levels are cumulative: someone operating at level 4 can still handle level-1 tasks in a minute.
The four levels, patent-specific
Occasional user
ChatGPT occasionally, a few times per month: rephrasing, concept explanation, a quality check on a text. No fixed structure, no duty of evidence.
Human core skill: awareness that the tool exists.
Next step: open the Prompting Library and try a prompt for a specific recurring task. For a structured start, walk through the Generative AI course.
Systematic user
Works with Context Engineering: every task gets a fresh chat with exactly the relevant documents. Duty of evidence is internalised, every AI statement is expected to come with a quote and a reference. Self-reflection at the end of every analysis.
Human core skill: self-discipline, structuring one's own steps consistently.
Next step: run a topic-specific prompt that enforces several steps: scope-of-protection prompt (15 steps) or office-action response prompt (5 steps).
Workflow user
Whole workflows run AI-supported: defined steps each with its own prompt, drafter/reviewer split on critical passages, devil's-advocate pass as standard. Context Engineering (context engineering) is already automated at this level — the workflow assembles the right documents per step, the attorney no longer curates a chat by hand. Quality standards are formalised, deviations are deliberately justified.
Human core skill: systems thinking, seeing the work as a process.
Next step: read the Agent Framework and deliberately apply multi-model workflows in your next critical mandate.
Tool maker
Builds own skills, prompts and automations for the firm. Data analysis, prior-art watching and research synthesis partly run automatically; the attorney reviews results rather than producing them. Trains colleagues and keeps the firm's AI governance current.
Human core skill: leadership, building and sustaining an AI culture.
Next step: install the patent plugin, write your first own skills in SKILL.md format and start a small n8n project, e.g. a prior-art watch. The Claude Code Guide explains the skill format.
30-second self-assessment
Answer the questions in order until you hit a "no". That is where you are now. The level above is your next step.
- Do I use AI for my work at least once a month? (Level 1)
- Do I start a fresh chat with exactly the relevant documents for each task and keep duty of evidence? (Level 2)
- Does at least one workflow run AI-supported end-to-end with a drafter/reviewer pass (e.g. office-action response)? (Level 3)
- Have I built at least one skill or automation of my own that colleagues also use? (Level 4)
Why the progression is not linear
Not every firm has to reach level 4. A boutique firm often benefits more from a clean level 2 than from a half-baked level 3. The question that matters is: which combination of levels fits my mandate portfolio and my team?
It is also typical that different workflows operate at different levels. Office-action response at level 3 with a drafter/reviewer pass, FTO at level 2 with Context Engineering and duty of evidence, drafting at level 3 with a formalised workflow and a local model. Every such combination is legitimate.
Connection to the other pillars
The levels build on the four pillars for prompt engineering in patent practice. From level 2 onwards they become standard equipment:
- Context Engineering: mandatory from level 2, automated by the workflow from level 3, agent-driven from level 4.
- Duty of evidence: mandatory from level 2, policed by the reviewer from level 3.
- Paragraph-by-paragraph search: mandatory from level 2, parallel between drafter and reviewer from level 3.
- Self-reflection + devil's advocate: self-reflection from level 2, a real second-opinion pass from level 3.
Newsletter
Stay up to date on generative AI in patent practice.
Join the GenAI Newsletter