Pillar 5 is the operating engine of the Legal AI OS. Strategy without execution remains slideware. Governance without execution remains policy. The other seven pillars exist to support what Pillar 5 produces: the selection of which AI work to do, the operating cadence under which it gets done, the autonomy gradient at which AI systems act, and the measurement frame that produces the institutional case for the next funding cycle.
Functions that build Pillar 5 well treat use-case selection as a disciplined portfolio decision, not a tool-driven shopping exercise. They calibrate materiality before they delegate. They measure return across four categories rather than one. They progress through the Agentic Tier ladder deliberately, with named gates between tiers. They produce the operating evidence that Pillars 4 and 7 then audit.
The four operating decisions Pillar 5 governs
5.1 Use-case selection
Pillar 5 begins with a portfolio of candidate use cases scored against the ROAI 4-Quadrant framework. Each candidate is scored on productivity gain (Q1), defensibility contribution (Q2), institutional standing (Q3), and category positioning (Q4). The portfolio is rebalanced quarterly. Use cases are not promoted on Q1 alone; functions that promote on productivity alone build a portfolio that loses to fuller-frame portfolios at the next funding cycle.
Selection discipline includes explicit rejection criteria. Use cases that cannot satisfy Pillar 4 evidence requirements are rejected at intake, not after pilot. Use cases that require autonomy levels the function cannot govern are rejected. Use cases that depend on data infrastructure the function does not yet have are deferred to a later cycle with Pillar 2 dependencies named.
5.2 Materiality calibration
Materiality calibration determines how consequential a given AI-influenced output is. Low-materiality work (formatting, deduplication, low-stakes summarisation) tolerates Tier 2 co-pilot operation with light review. High-materiality work (advice memos, regulatory positions, client-facing deliverables) requires either Tier 1 augmentation with full human authorship, or attested Tier 2 operation with documented partner sign-off. Calibration is a Pillar 5 discipline because it determines the tier at which each piece of AI work operates.
5.3 Tier progression on the Agentic ladder
The Agentic Tier ladder runs from Tier 1 (Augmentation) through Tier 2 (Co-pilot), Tier 3 (Workflow operator), to Tier 4 (Autonomous agent within bounded scope). Pillar 5 governs how the function progresses up the ladder. Each tier transition requires evidence of competent operation at the prior tier, a defined delegation-authority register, and a Pillar 4 governance review of the autonomy band.
Most institutional legal functions in 2026 sit at Tier 1 or low Tier 2 across most use cases. The progression to Tier 3 and bounded Tier 4 is not a tooling change. It is an institutional capability build that Pillar 5 paces deliberately.
5.4 Measurement against ROAI
Pillar 5 produces the measurement that Pillar 7 then aggregates. Each use case carries baseline metrics, defined target metrics across the four ROAI categories, and a quarterly review cadence. The measurement frame is built into the operating discipline rather than retrofitted at funding cycles. Functions that retrofit measurement at funding cycles produce defensive cases; functions that build measurement into operation produce institutional cases.
Common failure modes
Pillar 5 fails in four recognisable patterns. Tool-driven selection: use cases are chosen because a vendor pitched them rather than because the portfolio needed them. Productivity-only measurement: only Q1 is tracked, so the case the function presents to the board is the weakest of four possible cases. Autonomy creep: tier progressions happen informally as practitioners experiment, without governance review, producing latent defensibility exposure. Demonstration-grade execution: pilots are designed to look good in a deck rather than to operate at production cadence; the function ships pilots that never scale.
What success looks like at Bands 4 and 5
At Maturity Band 4 (Optimised), Pillar 5 produces a published portfolio with ROAI 4-Quadrant scoring, a delegation-authority register, quarterly measurement reviews, and tier progressions governed at a named cadence. At Band 5 (Defensible), the portfolio carries attestation that each promoted use case has cleared Pillar 4 evidence requirements, autonomy bands are documented per use case, and the function can produce a complete operating history of any AI-influenced decision on request.
Interlock with adjacent pillars
Pillar 5 depends on Pillar 1 for the strategic mandate that defines acceptable scope, Pillar 2 for the data and knowledge infrastructure each use case rests on, Pillar 3 for the literacy that practitioners require to operate AI competently at the chosen tier, Pillar 4 for the governance posture that bounds autonomy and produces evidence, Pillar 6 for vendor selection that matches use cases, Pillar 7 for the maturity progression that situates the portfolio, and Pillar 8 for the lifecycle discipline that retires use cases as the portfolio evolves. Pillar 5 is where the other seven pillars become visible as operating outcomes.