From hourly rates to AI agents: in-house legal rewrites the playbook

Legal teams feel the AI spark - here's how to turn it into repeatable results

Energy, finance, telecoms and tech legal departments are moving fast on AI. The trigger is real work: grid upgrades, digital bonds, cross-border compliance, and tighter budgets. The pattern is consistent - smaller teams, clearer workflows, and AI taking the first pass.

In the UK, National Grid is upgrading its network with more than £30bn of projects, adding legal load across 17 major transmission programmes. The legal team expanded headcount but set a limit and changed how work gets done: legal work management, AI-first drafting and review, and structured collaboration with panel firms.

Close to 100 National Grid lawyers now use the Harvey platform, reporting 3-5 hours saved each week today, with a target of a day per week. The team is also phasing out hourly rates by 2027 for most matters, running weekly joint calls with firms on its largest developments, and asking firms to co-build tech-enabled processes - with "many millions" allocated, but spent differently.

Accenture trained 3,300 legal professionals for "personal reinvention" in an AI-enabled workflow. The role is shifting to design, quality assurance, and relationships - with lawyers as the human in the loop who can override confident but wrong outputs. The team has built 35 AI agents (12 live), including one that summarises regulations across 50 US states.

Lloyds Banking Group reorganised into small squads with short-cycle goals, clearing blockers quickly and keeping momentum high. Other in-house teams report similar wins after integrating AI into intake, document management, and matter tracking.

What leading teams are actually shipping

• National Grid: Broad AI adoption, a digital committee with UK panel firms, and a clear stance on pricing and technology.
• Ilunion: Custom AI agents cut time on repetitive work by ~30% across contracts and case analysis.
• Repsol: New DMS and AI tools cut document review time by ~75%; more than half of workflows now automated.
• ASML: A chatbot that synthesises regulatory updates across panel firms to brief lawyers on export controls and other rules.
• UBS: Data-driven outside counsel framework cut big-project spend by ~25% using AI for benchmarking and staffing forecasts.
• Endava: AI hackathon produced tools that helped the team handle ~3x more business requests with faster turnaround.
• HSBC: Central AI committee guiding adoption and training; automated trademark licensing approvals.

A practical playbook for in-house legal

• Set a clear outcome per workflow: time saved, cycle time, quality bar, and risk thresholds.
• Fix intake first. Standardise forms, add triage, and route based on urgency, value and risk.
• Choose 2-3 high-volume use cases: contract review, clause comparisons, and knowledge search.
• Stand up AI governance: approved tools, data protection, IP rules, audit logs, and escalation paths.
• Train for prompting and review. Teach lawyers where AI is strong, where it fails, and how to test outputs.
• Adopt human-in-the-loop tiers: low, medium and high assurance workflows with clear sign-off rules.
• Build lightweight agents that chain tasks (intake → draft → review → summary) inside secure environments.
• Restructure work into small squads with time-boxed goals and visible blockers.
• Reset outside counsel terms: fewer firms, fixed or outcome-based fees, and shared tooling.
• Measure weekly. Publish usage, time saved, defect rates, and rework. Kill what doesn't move the needle.

Outside counsel: the new deal

• Phase out hourly rates on suitable work; use fixed fees, subscriptions, or matter-stage pricing.
• Require firms to use agreed tooling (playbooks, clause libraries, AI review) and contribute to joint knowledge bases.
• Hold a weekly cross-matter forum for your biggest programmes to surface common issues and speed decisions.
• Share data on quality and throughput; reallocate work based on performance, not legacy relationships.

AI agents worth building this year

• Regulatory summariser across jurisdictions with citations and confidence flags.
• Clause comparator that proposes playbook-approved edits and tracks supplier changes.
• Legal knowledge chatbot that answers on policy, templates, and matter history with links to sources.
• Self-serve NDA and low-risk contracts with auto-approval rules.
• Regulatory change tracker feeding alerts and checklists to matter owners.

Safeguards that keep you out of trouble

• Use enterprise-grade environments; restrict training on client data; apply data minimisation.
• Install retrieval-augmented generation and citation checks; require spot tests on key outputs.
• Run red-team reviews for bias, hallucinations, and leakage; log prompts and decisions.
• Teach confidence calibration: lawyers should feel able to reject plausible but wrong drafts.

Quick wins you can ship this quarter

• Automate intake and routing with capacity signals and SLAs.
• Add clause comparison and template variance reports to cut needless amendments.
• Stand up document review copilots inside your DMS with playbook prompts.
• Automate trademark or low-risk brand approvals with self-certification plus audit trails.

For a view on digital bonds under English law, see Euroclear's Digital Financial Market Infrastructure, which several banks and issuers have used for native issuance. Learn more.

If your team needs structured, practical upskilling on prompting and agent workflows, explore short courses by role here: Complete AI Training - Courses by Job.

Bottom line

The pattern is clear: simplify process, set guardrails, train for review, and make firms part of the system. The result is measurable time back, better quality, and fewer surprises - at scale.