GenAI in Law 2025: Copilot Jylo and the 2026 governance playbook

Tony McKenna’s reflections from Lawfront capture 2025 as the year generative AI moved from experiment to organizational imperative for many law firms — a year when Microsoft Copilot and specialist platforms such as Jylo shifted from pilots into daily workflows, and when firms began to wrestle publicly with the long-term choices that follow rapid GenAI adoption. The conversation is as much about technology as it is about governance, talent and vendor strategy: McKenna warns that while vendor-led copilots deliver immediate productivity, building and sustaining proprietary AI is a longer, costlier commitment — a trade-off the profession must confront as it plans for 2026.

Background / Overview​

2025 closed with clear momentum for legal technology adoption. Two structural shifts stand out: first, enterprise copilots embedded in core productivity suites made practical inroads because vendors relaxed licensing and purchase barriers; second, vertical legal AI platforms and marketplaces grew as firms sought provenance, customization and a way to package partner expertise into repeatable playbooks. Both trends are reflected in Lawfront’s choices — the firm doubled down on Microsoft technologies while also investing in Jylo to provide an internal marketplace for firm-specific AI playbooks. These are not isolated tactical choices; they reflect a broader market movement toward integrating GenAI into document-centric workflows, precedent search, contract analysis and matter triage. Microsoft’s commercial posture made a practical difference in 2025. The company removed aggressive seat minimums and extended partner promotions that widened Copilot availability for smaller tenants and CSP partners, enabling law firms of many sizes to buy into an integrated Copilot experience tied to Word, Outlook, SharePoint and Teams. Those licensing changes materially lowered the barrier to entry for firms that were previously locked out by minimum-seat economics. At the same time, Microsoft continued to expand Copilot features — including administrative Copilot surfaces and deeper SharePoint/Teams integration — making a Microsoft-centric stack a viable strategic backbone. Concurrently, specialist offerings like Jylo emerged as pragmatic alternatives for firms that want a marketplace-style approach to legal playbooks, model choice, and tenant-grounded indexing. Lawfront’s public selection of Jylo illustrates the value many firms place on a platform that can host multiple LLMs, package expert partner knowledge, and produce reusable, auditable playbooks for common tasks — from clause extraction to document review. That combination of tenant control and marketplace flexibility explains why midsize and regional firms see value in these vertical tools even as the big platforms expand their own copilot capabilities.

---

Why 2025 felt different: licensing, integration and the multi-tool reality​

Licensing and availability shifted incentives​

Two practical licensing moves in 2024–25 changed buying behavior:

• Microsoft removed the former 300-seat minimum and expanded Copilot availability to more SKU pathways, enabling broader procurement options.
• Partner promotions and expanded CSP caps reduced transactional friction for channel partners to bundle Copilot into smaller law firm deals.

These commercial shifts turned Copilot from a theoretical productivity boost into a purchasable, auditable product for many firms — particularly those already embedded in Microsoft 365. The result is an acceleration of vendor-led adoption: firms buy a Copilot-enabled productivity baseline and then layer specialist tools where needed.

Integration — the SharePoint and Teams vector​

Copilot’s value increases when it can access firm content with provenance. Integration into SharePoint, OneDrive and Teams — combined with semantic indexing — makes Copilot not just a drafting aid, but a knowledge surface that can pull from matter files, precedents and internal playbooks. Expect a big focus in 2026 on leveraging those integration points to create defensible, auditable AI outputs that lawyers can sign off on. Microsoft’s roadmap for Copilot admin experiences and semantic index capabilities shows this is not hypothetical: the platform is explicitly aimed at surfacing relevant, tenant-specific context into responses.

The multi-tool reality: why firms don’t pick a single winner​

Large law firms are continuing a pragmatic pattern: they combine enterprise copilots with specialist legal LLMs, RAG (retrieval-augmented generation) stacks, and tenant agents. The reason is practical — no single product yet covers every workflow and compliance posture:

• Copilots excel at embedded productivity tasks and meeting summarization.
• Legal-specialist models claim better accuracy for precedent and citation reasoning.
• RAG pipelines and private indexes provide provenance and explainability.
• Proprietary playbooks (e.g., Jylo-style marketplaces) enable reuse of firm-specific knowledge.

That multi-tool stack is powerful but introduces procurement and governance complexity, and it raises questions about operational support and long-term maintenance.

---

Tony McKenna’s pragmatic prescription: partner, don’t over-commit to DIY​

McKenna’s central advice to firms is cautionary and practical: building your own cathedral-scale GenAI stack is possible but requires sustained investment in engineering, data ops, security, and subject-matter model training. That commitment often exceeds the appetite and capacity of many firms, particularly at regional scale. Instead, McKenna advocates partnering with specialized vendors who:

• Understand legal workflows,
• Offer tenant-grounded model controls,
• Provide productized support and long-term upgrades,
• And deliver measurable, repeatable workflows that lawyers can trust.

This partner-first stance doesn’t reject internal innovation; it prioritizes external collaboration where vendor expertise and product roadmaps reduce risk and accelerate time-to-value. It’s a call to weigh total cost of ownership — not only initial development — before launching in-house model programs.

---

The promise and the peril: what worked in 2025 and what to watch in 2026​

Tangible wins in 2025​

• Drafting and first-draft acceleration — Firms reported marked time reductions for routine memos, letters and clause generation; Copilot and specialist copilots made initial drafts quicker to produce.
• Meeting and transcript summarization — Embedded copilots in Teams accelerated capture and extraction of action items and meeting highlights.
• Contract triage and clause extraction — RAG-enabled pipelines and playbook-based automation yielded real reductions in manual review time.
• Business resilience via multi-vendor stacks — Firms balanced convenience and accuracy by combining platform copilots with specialist legal AI tools.

These benefits translated to measurable throughput gains in trial deployments and gave firms confidence to scale pilots into matter-level tests when paired with human verification.

Material risks that demand investment​

• Hallucinations with professional risk — The legal profession is uniquely exposed to hallucination risk: fabricated citations or mischaracterised precedent can lead to sanctions and reputational damage. Courts and bar authorities are paying attention, and firms must ensure auditable, human-in-the-loop verification processes.
• Data leakage and retraining — Unclear vendor practices around data retention and model retraining can expose confidential matter data. Contracts need enforceable deletion and no‑train clauses, and technical controls must prevent matter text from being absorbed into vendor-general models.
• Deskilling and apprenticeship loss — If junior lawyers stop drafting and redlining as part of their training, firms risk hollowing essential legal judgement and writing skills. Designated training pathways and rotational programs are necessary to preserve skill formation.
• Vendor lock-in and exit fragility — Deep integration with a single copilot vendor creates switching costs. Firms must negotiate exit rights, data egress and observable model behavior before committing strategically.

These risks are manageable but costly; they require investment in policy, people and observability as much as in models or compute.

---

Strategic playbook for 2026: governance, procurement and people​

1. Governance and procurement: make AI procurement a legal workstream​

• Treat AI procurements like any other legal product with warranties, deletion clauses, and audit rights.
• Require SOC/ISO or equivalent attestations from vendors and demand event-level logs and model-version metadata for all matter-facing outputs.
• Negotiate no-train/no-retain and data-residency clauses where client confidentiality is at stake.

Legal teams must own procurement language early; IT cannot manage model risk alone. This is now a legal governance issue as much as a technical one.

2. Pilot to production with measurable gates​

• Define a narrow, high-volume pilot (e.g., NDA triage, first-draft memos).
• Require reproducible test datasets and baseline metrics.
• Measure not just time saved but verification cost — how many minutes of lawyer review per draft, and error rates requiring rework.
• Only graduate pilots that demonstrate durable accuracy and manageable verification overhead.

This pilot-to-gate discipline prevents the “pilot sprawl” that burned many early AI projects.

3. Invest in human verification and role redesign​

• Create roles such as AI verifiers, knowledge managers, and AgentOps leads who own prompt hygiene, verification playbooks, and RAG quality.
• Preserve training for junior lawyers — pair automation with supervised learning opportunities and formal competency gates.
• Treat human review as a measurable step in workflow cost accounting, not an afterthought.

The net effect: firms capture productivity while preserving professional standards and training pipelines.

4. Prefer partner-first approaches for most firms​

• For many regional and mid-market firms, partnering with vendors like Jylo or continuing to extend their Microsoft stack will give faster, lower-risk returns than building in-house large models.
• Where firms do pursue in-house models, treat them as long-term engineering projects with committed multi-year budgets for data ops, MLOps and security.

McKenna’s view is practical here: partner when partner product-market fit exists; build when you have a sustained, defensible need that vendors cannot or will not meet.

---

The Microsoft axis: hold on to your SharePoint connectors​

McKenna’s prediction for 2026 is a continued focus on Microsoft, especially SharePoint-driven experiences. This is a strategically sensible position for many firms: Microsoft’s copilot family is embedded in Word, Outlook and Teams — the core tools lawyers use — and the semantic indexing and tenant-grounded capabilities make it attractive for knowledge retrieval and contextual responses.
Two realities support that bet:

• Microsoft’s licensing and distribution changes made Copilot accessible to a broader range of organizations in 2025, encouraging firms to standardize on Microsoft stacks as a low-friction route to AI capability.
• The productivity uplift from embedded copilots is real when coupled with a semantic index and SharePoint as the canonical content store; firms that consolidate content into controlled repositories will extract the most value. However, this convenience must be balanced against vendor concentration risk and the need for contractual protections.

McKenna’s call to “hold on to your hats” captures the velocity of change: integration and feature rollouts will accelerate, and firms integrated into Microsoft’s ecosystem will likely see fast-moving new capabilities they’ll need to evaluate and govern.

---

Events, community momentum and the social proof effect​

2025 was a strong year for legal technology events — ILTA’s large conference in the U.S. national-harbor area and ILTA Europe’s London return reinforced community practices, vetting, and peer-to-peer learning. Those events served as practical forums where product narratives met real-world deployments, accelerating adoption by providing social proof, shared lessons on governance, and hands-on sessions. The ILTA calendar and attendance made it clear that legal tech professionals are rapidly professionalizing AI adoption practices and operational playbooks. Formal community moments — from vendor panels to governance workshops — materially reduce adoption risk because peers can compare metrics, share redlines for vendor contracts, and catalogue verification checklists. For 2026, expect the conference circuit to remain central to procurement decisions and to accelerate consensus around “acceptable” guardrails for AI in practice.

---

A note on unverifiable claims and cautionary language​

Several market narratives in 2025 — rapid adoption percentages, anecdotal productivity claims, and vendor ROI figures — are directional rather than precise. Where numbers are quoted publicly they often reflect vendor surveys, which may measure perceptions rather than independently audited time savings. Caution is warranted when applying headline productivity percentages to firm-wide budgeting without running local, representative pilots and measuring verification overhead. McKenna’s reflections are practical here: treat vendor claims as hypotheses to test, not guarantees. Similarly, some vendor announcements hinting at very large enterprise deployments (e.g., reported large Copilot deals) may not reflect final contractual terms, discounts, or feature parity across tenants. Firms should verify availability, commercial terms and compliance features directly with vendors before strategic commitments.

---

What 2026 will likely bring — predictions built on 2025 dynamics​

• Continued Microsoft centrism for firms already standardized on the Office stack — deeper SharePoint and Teams integration will be the quickest path to scalable, auditable GenAI outcomes.
• Rapid growth of vertical AI marketplaces and playbook platforms (Jylo-style) as firms prioritize reuse, auditable workflows and cross-firm sharing of expertise. Expect more consolidation and larger partnerships between platform vendors and legal groups.
• A market bifurcation: vendor-led copilots for mainstream productivity vs. bespoke in-house models for firms with unique IP — the latter will remain rare and capital-intensive.
• Intensifying regulatory and bar attention to provenance — audio, document and model logs will become expected evidence in disputes and professional reviews. Expect formal guidance and possibly reporting requirements in several jurisdictions.
• Vendor consolidation and a sharper buyer market: only vendors with demonstrable, reproducible accuracy and contractual protections will persist; expect M&A and shakeouts as investors refocus portfolios.

---

Final analysis — strengths, blind spots and practical counsel​

Tony McKenna’s reflections are notable for pragmatic restraint: he celebrates the community momentum and the operational wins of 2025 while warning against naive, short-term thinking about building proprietary GenAI. His preference for partnering with vendors that understand legal workflows is a strategic, risk-aware stance that many firms — especially regional consolidators like Lawfront — will find sensible.
Strengths:

• Clear recognition of the operational value of integrated copilots and vertical playbooks.
• Practical governance-first advice that treats procurement as a legal and operational problem, not merely an IT one.
• A realistic view of the long-term costs of building proprietary models and systems.

Blind spots and risks:

• The pace of feature rollouts from major vendors can outstrip a firm’s governance readiness; change management is a system-level requirement, not a checkbox.
• Productivity percentages from vendors should be treated as directional; rigorous local validation is required before repricing or headcount decisions.
• Market consolidation could entrench platform dependency; active exit planning and data portability requirements are essential.

Practical counsel for 2026 execution:

• Make AI procurement a legal-led process with enforceable data and training protections.
• Pilot narrowly and measure both speed gains and verification costs before scaling.
• Invest in human verification roles and structured apprenticeship pathways to protect professional standards.
• Prefer vendor partnerships for broad capability; reserve in-house build for narrow, defensible IP where long-term investment is justified.

---

In sum, 2025 was the year GenAI stopped being merely hypothetical for many legal teams and became an operational decision with real costs, benefits and professional consequences. McKenna’s reflections — from Lawfront’s Microsoft focus to its choice of Jylo for AI playbooks — illustrate a pragmatic, partnership-led route forward. The law firms that will succeed in 2026 are not necessarily those who build the largest models, but those who combine sound governance, disciplined procurement, human verification and the right vendor partnerships to turn AI’s raw potential into durable client value.

---

Source: Legal IT Insider 2025/26 Reflections & Predictions: Tony McKenna, Lawfront - A significant increase in GenAI adoption - Legal IT Insider