The image of a junior lawyer working through stacks of contracts late into the night, manually building risk lists, has not fundamentally changed at many Swiss law firms. But the pressure is building – and it is coming from several directions at once. AI tools are taking over tasks that once made up the bulk of an associate's training time. At the same time, the revised Swiss Federal Act on Data Protection (revFADP / revDSG), in force since 1 September 2023, has introduced new requirements for handling personal data in client-related contexts. And with the Federal Council's ongoing consultation on AI regulation (as of Q1 2025), the regulatory framework for AI use in legal practice is likely to become more concrete in the near term.
Law firms that do not adapt associate training now are educating lawyers for work that will rarely need to be done manually within a few years – and leaving aside the new competencies that are genuinely in demand.
This is not a question of enthusiasm for technology. It is about which skills are genuinely needed today – and which ones still require real human mastery because AI cannot reliably replace them.
What has already changed in contract work
Contract review, risk analysis, and checking standard clauses have traditionally been core tasks for associates in their early years. This is precisely where legal AI is making inroads. Platforms like CASUS, a Swiss legal AI platform with hosting in Switzerland and the EU, can review a contract for risks and red flags, prioritize findings by severity (low / medium / high), and deliver drafting suggestions directly in Word – without copy-paste and with correct formatting.
A concrete example: a Zurich M&A team used CASUS's Benchmark module to review a 120-page share purchase agreement. Checking for missing standard clauses – liability cap, warranty catalogue, MAC definition, completeness of W&I provisions – previously took around four hours on first pass. With the Benchmark module, a prioritized gap analysis was available within approximately 45 minutes, including a percentage match against the firm's internal playbook and specific insertion suggestions for missing clauses placed correctly within the document. The associate could move directly to evaluating identified gaps rather than locating them.
That does not mean associates no longer need to understand these tasks. It means they need to learn how to assess, refine, and take responsibility for AI outputs. Anyone who does not know how to evaluate a liability clause without a cap under Art. 100 OR cannot correctly interpret an AI finding either.
The shift is less about substitution than about reallocation – away from mechanical first-pass review, toward qualified legal judgment.
Which skills are genuinely needed now
Critical thinking about AI outputs
The most important skill first: AI tools produce structured, source-based results – but not guaranteed correct ones. Associates need to learn how to systematically question AI outputs. Does the risk assessment align with the actual legal position? Is the cited case law relevant? Is something missing because it was never prompted?
In practice, two mistakes come up repeatedly when associates are first working with legal AI. First, severity flags (high / medium / low) get too much weight without checking whether a finding is actually relevant in the specific negotiation context – a clause flagged as "high" severity may have been accepted deliberately in a given deal. Second, alerts about missing definitions get skipped systematically, even though definition gaps in long contracts are among the most expensive disputes that arise. The Swiss Federal Supreme Court's ruling BGE 144 III 327, addressing contract interpretation under Art. 18 OR, illustrates how undefined or ambiguous terms create interpretive disputes that careful drafting could have prevented.
Critical thinking toward AI is not a philosophical exercise. It is a practical legal skill expressed in concrete review steps: source check, completeness control, plausibility test.
Prompt skills as a legal work technique
A vague question put to an AI system produces a vague answer. Well-structured prompts – with context, the specific question, party position, and desired output format – produce usable results. That sounds like a technical skill, but at its practical level it is legal precision: the same ability needed to write clear pleadings and contract clauses.
Associates who learn how to formulate a legal research prompt for a liability question under Art. 97 ff. OR are simultaneously learning how to structure the legal question itself. A weak prompt might return a general overview of contractual liability. A precise prompt – specifying the facts, party position, and desired output structure – can return a structured assessment with risk drivers, argument lines, and a concrete recommendation that feeds directly into an internal memo.
Prompt engineering and legal thinking are not opposites here.
Contract and document review with AI support
The ability to conduct and take responsibility for an AI-assisted contract review is becoming a standard competency. Concretely: working with the Risk & Quality Review module, evaluating findings, selectively accepting suggestions, and being able to explain the analysis to a client. Anyone who knows the workflow – from risk analysis to improvement recommendation – can work more efficiently and better articulate what was reviewed and what was not.
The same applies to the Benchmark workflow: checking an NDA or SPA against an internal standard and assessing deviations is a competency that previously required significant experience. With structured AI support, that judgment can be developed earlier in a career – provided the legal ability to interpret results is in place.
Legal research with AI databases
Legal research is changing fundamentally. What previously meant hours in a database can now mean a well-formulated prompt across more than 660,000 cantonal and federal court decisions, with the relevant reasoning sections visible directly in the search results – no need to open each decision individually.
CASUS's legal research module produces structured first assessments with risk drivers, pro and con argument lines, and concrete recommendations. For associates, this means the ability to produce a source-based first assessment arrives earlier in their careers. But the ability to place that assessment in legal context, weigh it critically, and translate it into a work product remains human.
Formal and linguistic document quality
AI also handles proofreading tasks: spelling, grammar, consistency of terminology, cross-references, missing definitions, placeholders. The Proofread module checks Swiss spelling conventions (ss instead of ss), annexes, numbering, and flags contradictions – taking work off the table that is necessary but time-consuming.
The goal is not a legal judgment on the legal situation, but a clean, consistent document. Associates need to understand this distinction too: what AI proofreading covers and where it ends.
Competency matrix: traditional training vs. the AI era
The table below makes the shift concrete – not as replacement, but as a recalibration of priorities. It also distinguishes between year-1 associates who are first working with legal AI, and year-3 associates who are beginning to own workflows.
Competency area | Traditional training | AI era – Year 1 | AI era – Year 3 |
|---|---|---|---|
First-pass / red-flag analysis | Manual, time-intensive, core task | Review AI output, evaluate findings | Design review process, quality control |
Case law research | Database search, manual filtering | Structure prompts, interpret results | Standardize research workflows |
Clause negotiation | Experience built through repetition | Spot deviations from Benchmark output | Maintain playbook, adjust AI standard |
Document quality | Manual proofreading | Verify AI proofreading output | Define firm-wide consistency standards |
Data protection competency | Basic DSG knowledge | Apply revFADP Art. 5 / Art. 22 to AI workflows | Evaluate and approve AI tools under revFADP |
Client communication | Reports prepared manually | Contextualize and approve AI drafts | Explain efficiency gains to clients |
Process design | Little role at junior level | Document own prompts and workflows | Legal engineering: shape firm processes |
This table is not an ideal curriculum – it is an orientation framework. Each firm will weight the areas differently depending on practice group and client structure. What matters is that the shifts are named explicitly, so that associates are not implicitly trained for a competency profile that no longer matches current demands.
How firms can adapt associate training
Structured introduction to AI workflows
Simply providing an AI tool without onboarding achieves little. Associates need clear explanations of what each module is suited for, where its limits lie, and how to interpret outputs. A structured introduction to the review workflow – using a real contract the firm knows – delivers more than an onboarding document.
A format that works well in practice follows three steps: first, see the AI output without any prior analysis; then form your own legal assessment; then compare the two. This comparison quickly reveals where associates over-rely on AI findings – for instance, accepting a "high" severity flag without checking whether the underlying clause was intentionally negotiated that way in the specific deal context.
Case-based learning with AI as a benchmark
There is a concern that case-based learning loses relevance once AI takes over first-pass analysis. This is only partly true. Assessing an AI output requires understanding the subject matter. Someone who has never learned to read a liability clause and place it in the context of Art. 100 OR cannot judge whether an AI finding is correct.
From onboarding observations, two patterns come up repeatedly: associates accept severity flags without considering negotiation context, and they systematically skip missing-definition alerts – even though definition gaps are among the most costly disputes in long-form contracts. Case-based training that addresses these specific failure modes directly is more effective than general guidance on AI use.
Associates should form their own hypothesis before seeing the AI result, then evaluate what the AI identified correctly and where it was wrong or incomplete. That structure turns AI into a learning tool rather than a shortcut.
Data protection and security as a mandatory training element
The revFADP (Art. 5 lit. g) introduced the concept of automated individual decision-making into Swiss data protection law. Art. 22 revFADP governs high-risk profiling and sets requirements for transparency and legal basis. For associates using AI tools in client-related contexts, this is concrete: they need to know whether and how the systems they use process personal data, and whether that processing is compliant with revFADP.
This is not an abstract compliance question. Uploading a contract containing personal data to an AI system that transfers data to the US or uses documents for model training risks not only a data protection breach but also a breach of the professional duty of care owed to the client. CASUS transfers no data to the US, operates with no human review and zero data retention – these are relevant criteria for revFADP-compliant use in law firm contexts.
The EU AI Act (Art. 6) classifies certain AI applications as high-risk. For Swiss firms working for international clients or with EU counterparties, this is not purely a European concern: AI tools used in legal advice may fall under high-risk categories depending on the context of use, triggering transparency and documentation obligations. This belongs in associate training, not just in IT onboarding.
More on CASUS's hosting and security architecture at Security & data residency.
From legal training to legal engineering
The term "legal engineer" is not yet standard in Swiss job postings. But it describes a direction that is taking shape: lawyers who think legally and can also design processes, deploy tools, and structure outputs in ways that go beyond traditional practice.
This is not a departure from legal substance. On the contrary – taking responsibility for AI outputs requires solid legal foundations that go deeper than keyword familiarity. Art. 18 OR on contract interpretation, Art. 97 ff. OR on liability, the revFADP requirements on automated processing: these are not peripheral topics but the foundations on which every critical evaluation of an AI finding rests.
Legal engineering in practice means: documenting your own prompt workflows, feeding findings from the Benchmark module systematically back into the firm's playbook, and being able to explain efficiency gains – for example in SPA review – transparently to clients. That is a competency that extends well beyond the technical operation of a tool.
Associate training in the AI age means learning both. Legal substance and the ability to apply and take responsibility for it using structured tools.
Try CASUS in practice
Anyone who wants to see what AI-assisted contract work looks like in daily practice can test CASUS for free. The platform runs as a web app and as a Microsoft Word add-in – no data transfer to the US, with hosting in Switzerland and the EU, no human review, and zero data retention.
FAQ
What do associates need to learn differently in the AI age?
Associates primarily need to learn how to critically evaluate AI outputs: are the identified risks correct? Is something missing? Prompt skills – the ability to instruct AI systems precisely – become a regular work technique. Foundational legal knowledge remains indispensable, because it underpins every assessment of AI results. In practice, two failure modes come up repeatedly: accepting severity flags without checking deal context, and ignoring missing-definition alerts despite the legal exposure they can create.
Does AI replace the traditional associate training path?
No. AI takes over mechanical first-pass reviews and structured analysis, but not legal judgment. Training needs to adapt – away from manual first-pass review, toward the competency to take responsibility for AI outputs and work with them further. Foundational legal knowledge – for example on contract interpretation under Art. 18 OR or liability allocation under Art. 97 ff. OR – is a prerequisite, not an optional complement.
Which AI skills are most relevant for lawyers?
Practically relevant skills include: formulating prompts for legal tasks, evaluating AI-assisted risk analyses, using legal research tools with case law databases (over 660,000 decisions in CASUS), and understanding where AI proofreading ends and legal review begins. Added to that is knowing which AI systems may be used in a data-compliant way – a requirement that revFADP, in force since September 2023, makes concrete.
What does AI-assisted legal research look like?
Platforms like CASUS search across more than 660,000 cantonal and federal court decisions and produce structured assessments with source references, argument lines, and recommendations. The relevant reasoning sections are shown directly in the search result, without having to open each decision manually. That accelerates first assessment – but placing the result in legal context, for example in relation to a ruling like BGE 144 III 327 on contract interpretation, remains the lawyer's job.
How should law firms adjust associate training?
A structured introduction to specific AI workflows – not just tool access – is advisable, alongside case-based learning that uses AI as a benchmark rather than as the first analyst. Explicit training on data protection questions around AI use is also necessary, particularly on revFADP Art. 5 and Art. 22 and on the implications of EU AI Act Art. 6 for cross-border mandates. The competency matrix in this article provides an orientation framework for year-1 and year-3 targets.
Why is data security relevant in associate AI training?
Associates work with client-related, often confidential documents. They need to know which AI systems may be used in a data-compliant way. The revFADP (in force since 1 September 2023) brings concrete requirements for processing personal data, including in AI contexts – Art. 5 lit. g on automated processing and Art. 22 on high-risk profiling are particularly relevant. Systems that transfer data to the US or use documents for model training are not permissible in many client-related contexts.
What is the difference between AI proofreading and legal document review?
AI proofreading checks language, consistency, formatting, cross-references, and placeholders – for example whether a reference to section 7.2 actually exists, or whether a term is defined but used with a different spelling elsewhere. It does not assess whether a clause is legally correct or appropriate. A legal document review evaluates the legal position, risk allocation, and contract structure. Both functions complement each other – they do not substitute for one another.
Does this apply to in-house legal teams as well?
Yes. The shift in competencies affects in-house legal teams just as much as law firms – often more so, because capacity is tighter and contract volumes are high. A Basel-based pharmaceutical company with an international supplier portfolio reviewing dozens of NDAs and framework agreements each month benefits directly when its legal team can work systematically with the Risk & Quality Review module and the AI Data Room. The prerequisite is the same as in a law firm: the team needs to be able to place and take responsibility for the outputs.
What does the Federal Council's AI regulation consultation mean for law firms?
The Federal Council launched a consultation on AI regulation in Q1 2025, oriented toward a risk-based approach comparable to the EU AI Act. For law firms, this means AI tools used in legal advice may in the future be subject to transparency, documentation, and risk classification requirements. Associates who learn today how to evaluate and document AI outputs are better prepared for this regulatory direction than those who treat AI tools as black boxes.
