In the hallowed halls of traditional legal work, there remains a peculiar affection for the archaic: email ping-pong that stretches over fortnights, folders that multiply without explanation and, in the most determined corners, the occasional whirr of a fax machine. Charming in a museum. Expensive in a modern business.
The cost of inefficiency: a very expensive muddle
The trouble is rarely a complete absence of technology. Most legal departments already have document repositories, email, collaboration tools, contract databases and a collection of AI pilots. The muddle appears between them. Information is copied by hand, approvals disappear into inboxes and nobody can reconstruct why a decision was made.
This creates two equally unhelpful reactions. The first is the magic-wand assumption: buy an LLM and three Associates will somehow vanish from the cost base. The second is reflexive cynicism: dismiss Natural Language Processing as a glorified spellcheck while competitors use it to accelerate repetitive review and drafting.
Both miss the real problem. AI-driven efficiency is not a property of the model alone. It comes from the workflow around the model: the data it may see, the tools it may use, the rules it must follow and the human who remains accountable.
Shadow AI is a workflow symptom
If the approved route is too slow or too awkward, people will find an easier one. The answer is not merely another policy. It is a usable, governed alternative that fits the work people are trying to complete.
The rise of the Legal Engineer: the bilingual hero
The most effective legal teams do not simply hire better lawyers or better developers. They create a translation layer between the two. A Legal Engineer can turn a phrase such as “escalate material deviations” into an explicit definition of materiality, a set of data fields, a routing rule and an auditable approval step.
This does not require every lawyer to write Python. It does require legal teams to understand enough about LLMs for lawyers to ask the right questions:
- Which source is authoritative when documents disagree?
- What may the model read, and what may it change?
- Which decisions can be automated, and which require approval?
- How can a reviewer see the source, reasoning and action history?
- What happens when confidence is low or the case is genuinely new?
What Agentic AI changes
A conventional chatbot responds to a prompt. Agentic AI can coordinate a sequence of steps: find the relevant contract, retrieve a confirmed rule, compare the clause, draft an alternative, request approval and update the system of record. Standards such as Model Context Protocol can expose tools and resources to AI applications through a common architecture.
That does not make an agent autonomous in the legal sense. It makes the system capable of acting within the permissions and controls designed around it. A robust workflow therefore distinguishes between read access, proposed actions and confirmed write actions.
| Layer | Question the Legal Engineer answers |
|---|---|
| Model | Which task is the model good enough to perform? |
| Knowledge | Which rules, clauses and precedents may it rely on? |
| Tools | Which systems may it read from or write to? |
| Workflow | When does the matter proceed, pause or reach Legal? |
| Evidence | How are sources, changes and approvals recorded? |
Legal Tech compliance must be designed in
Compliance is not a disclaimer placed beneath a prompt box. The EU AI Act, data protection, confidentiality duties and internal governance influence the architecture itself. Access control, data minimisation, retention, human oversight and auditability must be translated into system behaviour.
The same applies to quality. A model may produce a convincing answer without a reliable basis. Legal workflows need visible sources, controlled rules and escalation paths for uncertain or high-risk cases. Where the system proposes rather than decides, the interface should make that distinction obvious.
Strategy and engineering belong in the same delivery team.
Our founders developed and implemented Legal Tech solutions in their former roles at Freshfields. At Boutiq AI, we combine that experience with product and engineering delivery. Treasure Contracts turns static contract files into usable process data and connects review, drafting, approval and signature without forcing every user into another system.
With Treasure Contracts, confirmed company rules can guide first-pass reviews, deviations are flagged as findings and passed to Legal through defined approval steps and every relevant step can be audited. Integrations bring the workflow into the CRM (HubSpot), into Slack and, via MCP, into AI assistants. The aim is not AI for AI’s sake. It is faster contract completion without Legal losing control.
Where a legal team should start
- Choose one high-volume workflow with a visible delay or error rate.
- Map the authoritative sources, decisions and responsible roles.
- Separate low-risk routine from matters that require legal judgement.
- Define permissions, approvals and evidence before adding automation.
- Measure cycle time, quality, adoption and exception volume.
The competitive edge will not belong to the team with the longest list of AI tools. It will belong to the team that can translate legal judgement into a secure operating model. Future-proofing legal work is less about finding a magician and more about building the bridge.
Sources and further reading
Published 17 August 2026. This article provides general information and does not constitute legal advice.