Legal AI: capability, build choices, and scenarios.

Findings from the May 2026 20Minds roundtable survey on AI in legal practice. Respondents rated confidence in four capability dimensions, ranked the case for and against building in-house tools, and assessed the plausibility of four scenarios for the future of legal work.

Respondents

Completed responses

Most confidence

7.8

Lawyer–AI collaboration

Least confidence

5.4

Confidentiality

Most plausible scenario

3.9

In-house context wins

Agenda & rules

Structure for the roundtable session. Three discussion blocks of 15 minutes each, followed by open sharing.

Focus of this session

LLMs are entering the legal space along several routes: integrated into legal-specific platforms (Harvey, Legora, Beamon), as plugins to foundational model providers (Claude for Legal), and in direct competition with established SaaS providers (Luminance). The question for legal teams: is this a real opportunity to automate more specific workflows, or is DIY a distraction from the core business of legal advice?

15 min

1. Introduction & capability confidence

Whether AI systems will, within a 2–3 year horizon, meet the threshold for material efficiency gains across four dimensions: hallucinations, confidentiality, lawyer–AI collaboration, and accountable outputs.

15 min

2. Scenarios

Four scenarios for how legal demand may shift over the coming years — from an "error storm" that increases work, to automated risk management that reduces it, to a re-allocation between in-house and external counsel.

15 min

3. Buy vs. build

The case for and against building legal AI tools in-house — flexibility and control on one side, talent and service responsibility on the other.

up to 10 min

4. Free sharing

Recommendations on providers, events, know-how, and other resources.

Rules

Chatham House Rules apply.
Introduction in round-robin format.
Use the "raise hand" function on Teams to indicate interest to speak.
A timer is used to distribute speaking time equally.

Confidence in AI capability

How confident are respondents that AI systems will, within a 2–3 year horizon, meet the threshold for material efficiency gains (40%+) across four dimensions? Scored 1–10. Click a row to see the full response distribution.

Four scenarios for the future of legal work

Respondents rated each scenario for plausibility on a 1–5 scale. Click a card to read the full scenario description and see the response distribution.

Buy versus build

Respondents rated their confidence in making buy-or-build decisions for legal AI tools (mean 7.0/10), then ranked five reasons for and five against building in-house. Bars show Borda-count totals: each first-place vote contributes 5 points, second place 4, and so on.

▸Reasons to build

▸Reasons not to build

Open comments

Verbatim comments submitted alongside the capability ratings and scenario ratings.

On capability and confidence

On the scenarios

Methodology

Sample. Eighteen completed responses from the 20Minds roundtable survey, collected via Typeform in May 2026.

Scoring. Capability and buy-or-build confidence questions used a 1–10 scale, framed as "today, with a 2–3 year horizon, averaged across all legal use cases." Scenario questions used a 1–5 plausibility scale (1 = implausible; 5 = fully plausible).

Rankings. Respondents ranked five options for and five against building in-house AI tools from most important (1) to least important (5). Rankings are aggregated using a Borda count: first place receives 5 points, second 4, third 3, fourth 2, fifth 1.

Material efficiency gains. The 40%+ threshold is the working definition the roundtable adopted for "material" rather than "incremental" productivity improvement.

Caveats. Small sample. Findings should be read as indicative of roundtable participant sentiment, not as a representative survey of the legal profession.