The previous essays in this series have set out, separately, the architecture, the data discipline, the safety framework, the deployment options, and the pricing structure that constitute the system. This essay is concerned with what the system, taken as a whole, actually does. It is a tour of the skill catalogue: two hundred and fifteen skills, organised across twenty-four functional groups, calibrated for fifteen distinct customer segments, and composable into team configurations that span the principal areas of legal practice. The tour is not exhaustive. It is structural. It aims to give the reader a feeling for the shape of the catalogue, the principles by which it is organised, and the kinds of work that each cluster of skills performs.
The catalogue is named MIRA, an abbreviation of Machine Intelligence and Reasoning Assistant. The name was chosen for what it indicates rather than what it sounds like. The system is a machine, in the sense that it is built and operated. It performs intelligence, in the sense that it produces reasoned outputs against legal questions. It performs reasoning, in the structured sense developed in our other writing, with chains that can be inspected. It is an assistant, in the sense that it operates under the supervision of a human professional and produces work product that the professional reviews, refines, and signs.
Two hundred and fifteen skills is a number that requires unpacking. It is large enough to span the working life of a substantial law firm, an in-house team, a regulatory body, or a judicial institution. It is small enough that the catalogue remains coherent, with each skill knowable, evaluable, and improvable. It is the result of deliberate growth rather than accumulation. The number will rise, but it will rise the way the catalogue has been built so far: skill by skill, each added because there is real work it performs, each evaluated against real standards, each integrated into the architecture rather than appended to it.
The catalogue is organised, at the highest level, by the kind of work performed. The principal organising dimensions are two: the functional group, which describes what kind of operation the skill performs, and the customer segment, which describes the audience for whom the skill has been calibrated. The two dimensions intersect to produce a matrix in which each cell may contain one or more skills. Some cells are densely populated; others are sparser; the overall shape reflects where the work has been most pressing.
The functional groups, of which there are twenty-four, span the operations that legal work consists of. They include groups for research, for drafting, for review, for analysis, for chronology and bundling, for compliance, for tax, for litigation strategy, for transactional support, for due diligence, for translation, for document comparison, for contract management, for regulatory tracking, for judicial preparation, for corporate governance, for competition, for employment, for real estate, for startup advisory, for litigation finance, for investigations, for the not-for-profit sector, and for legal education. Within each group, the skills are specialised by the type of output produced and by the standard of evaluation applied. A research skill in the litigation group is calibrated differently from a research skill in the corporate governance group, even though both perform research.
The customer segments, of which there are fifteen, span the kinds of institution in which the skills are deployed. They include large corporate law firms, mid-market firms, in-house departments at the largest companies, in-house teams in financial services, in-house teams in pharmaceuticals and life sciences, in-house teams in technology, in-house teams in manufacturing, government legal cells, judicial institutions, regulatory bodies, tax practices, chartered accountancy firms with legal functions, boutique litigation practices, small and growing practices, and academic institutions. Each segment has its own typical workload, its own user expectations, its own deployment patterns, and its own success criteria. Skills are tagged with the segments for which they have been calibrated, and configurations are built within segment-appropriate skill sets.
The intersection of group and segment produces a matrix that is, at any time, partially filled. The fill rate is highest in the groups and segments where the system has been most heavily deployed. It is lower in groups and segments where deployment is more recent. The catalogue grows by filling cells as the work demands.
A second classification of the skills, orthogonal to the group and segment, is by the kind of interaction the skill conducts with the user. The catalogue today contains 195 output skills, 19 hybrid skills, and 1 conversational skill. The classification is worth explaining, because it distinguishes the principal mode of the system from the supporting modes.
An output skill produces a defined artefact. The user supplies the inputs, the skill produces the artefact, the user reviews the artefact. The interaction is, on the output skill side, primarily one-way: the skill receives instructions and delivers results. The artefact is the deliverable; the conversation around it, where it occurs, is the user's review and the user's refinements. The 195 output skills in the catalogue cover the great majority of the working operations of the system. They include drafting skills, analysis skills, research skills, review skills, comparison skills, classification skills, summary skills, and the many specialised production skills that legal work consists of.
A hybrid skill combines output with structured interaction. The skill produces an artefact, but the artefact is built through a guided dialogue with the user. The user is asked to supply specific inputs at specific points; the skill produces intermediate outputs that the user reviews; the user supplies further inputs that refine the result. The 19 hybrid skills in the catalogue cover the kinds of work in which the user's expertise is integral to the production of the artefact, such as the structuring of a complex argument or the configuration of a transaction. The hybrid skills are designed for working sessions in which the user and the system iterate together, rather than for one-shot deliveries.
The single conversational skill is the open-ended dialogue interface, available for queries that do not naturally fit into a defined output or hybrid pattern. The conversational skill is the mode of the system that resembles a chatbot most closely. We provide it because it is, for some questions, the appropriate interface. We restrict it to a single skill in the catalogue, deliberately, because the architecture of the system is otherwise organised around defined skills, and we believe the user is best served by being directed to defined skills wherever they exist. The conversational skill is the fallback rather than the default.
The classification matters because it shapes the user's expectation. A user invoking an output skill expects an artefact and supplies the inputs the artefact requires. A user invoking a hybrid skill expects to participate in the production of the artefact and is prepared to do so. A user invoking the conversational skill expects an open dialogue. Each interaction is calibrated to the kind of work being performed, and the catalogue's mix of skill types reflects the structural reality that most legal work is best performed against defined deliverables.
Among the most heavily used clusters in the catalogue is the cluster of skills that support litigation practice. Litigation is the area in which the daily working life of a lawyer is most intensively shaped by the discipline of authority, the structure of pleadings, and the handling of large volumes of documentary material. Each of these aspects is supported by skills designed for the litigation context.
The research skills in the litigation cluster operate against the case law corpus, the statutory framework, and the procedural rules of the relevant forum. They identify controlling authorities, distinguish them by jurisdiction and currency, and produce the framework against which a position will be argued. The drafting skills produce pleadings, written submissions, applications, and the supporting documents that litigation requires. The pleading skills are calibrated by the procedural posture: an interlocutory application is drafted differently from a final argument, and the catalogue distinguishes accordingly. The review skills examine documents produced by opposing counsel, identify the points to be addressed, and structure the response.
The chronology and bundling skills, which are sometimes overlooked in discussions of legal AI but are essential to the operation of any large litigation practice, deserve their own mention. A chronology in a complex matter may run to hundreds of events. The construction of the chronology by hand is laborious; the maintenance of it as the matter develops is more so. The chronology skill in the catalogue ingests the available documents, extracts the dated events, organises them in chronological order, distinguishes the events that move the analysis from those that do not, and produces a chronology that is ready for use in the matter. The bundling skill performs the related but distinct operation of assembling the documents that will be relied on in court, in the format the court requires, with the index and the cross-references that the bundle conventionally carries. Both skills are unglamorous and high-leverage. A litigation team that uses them well saves substantial time on every matter.
The litigation cluster is, in our deployments, often configured as a litigation pod: a team of agents calibrated to work together on litigation matters, with the research, drafting, review, chronology, and bundling skills composed into a coherent workflow. A user supervising the pod is, in effect, supervising a junior team that handles the routine production of the matter, freeing the user to concentrate on strategy, authority selection, and the senior judgment that the matter requires.
A second heavily used cluster supports transactional practice. Transactional work has its own rhythms and its own demands. The volume of documents is high. The standards of consistency are exacting. The deadlines are tight. The integration of contributions from multiple parties, including counterparty counsel, is essential. Each of these aspects is supported by skills in the catalogue.
The drafting skills in the transactional cluster produce contracts, term sheets, side letters, and the related instruments that transactions require. The drafting skills are configured around the type of transaction: a share purchase agreement is drafted differently from a loan agreement, and the catalogue is structured accordingly. The review skills examine drafts produced by counterparty counsel, identify deviations from the firm's standard positions, and produce mark-ups that the user can refine. The clause comparison skills examine multiple agreements side by side, identify the variations in clause language, and produce a comparison that supports the user's evaluation of the alternatives.
The due diligence skills, central to most substantial transactions, are particularly worth describing. A due diligence exercise involves the review of large volumes of documentary material against a list of points to be checked. The traditional approach assigns the review to junior associates, who work through the documents one at a time. The catalogue's due diligence skills perform the routine portions of this work at scale, identifying the relevant documents, extracting the relevant terms, and surfacing the points that warrant senior attention. The user reviews the surfaced points rather than the raw material, with the senior attention concentrated on the matters that require it. The transformation of due diligence economics is, in our experience, one of the most consequential effects of careful skill deployment.
The transactional cluster is often configured as a transaction desk: a team of agents calibrated for the type of transaction at hand, with the drafting, review, comparison, and due diligence skills composed into a coordinated workflow. The desk operates against the firm's standard positions, produces drafts and reviews that are consistent with those positions, and surfaces the points that require negotiation. The user supervising the desk concentrates on negotiation, strategy, and the senior judgment that the transaction requires.
The cluster of skills that supports regulatory and compliance work is one of the largest in the catalogue, reflecting the scope of regulatory activity in contemporary practice. Regulatory work spans many areas: corporate governance, financial regulation, competition, data protection, employment, environmental, and the many sector-specific regimes that apply to particular industries. Each is supported by skills calibrated for the regulatory framework in question.
The framework tracking skills monitor the regulatory environment for changes that affect the user's matters. New notifications, amended rules, fresh circulars, recent decisions of regulatory authorities, all are surfaced in a structured form, with their implications for the user's operations identified. The advisory skills produce written advice on regulatory questions, calibrated to the audience for the advice, whether internal management, the board, or external regulators. The filing skills produce the regulatory filings that compliance work requires, in the format the relevant regulator demands.
The compliance review skills, designed for ongoing compliance work rather than for advisory matters, operate against the firm's existing compliance framework. They examine the firm's policies, procedures, and operations against the applicable regulatory requirements and identify gaps, weaknesses, and points where additional documentation or action is required. The compliance review is a rolling activity rather than a one-off, and the skills are designed to be invoked on a recurring basis as the regulatory environment evolves.
The cluster is often configured as a compliance factory: a team of agents calibrated for the firm's regulatory exposure, with the tracking, advisory, filing, and review skills composed into a continuous workflow. The factory operates against the firm's specific regulatory profile, produces the deliverables that compliance work requires, and surfaces the points that warrant senior attention. The user supervising the factory directs the work, applies senior judgment to the surfaced points, and ensures that the firm's compliance posture is maintained.
A further cluster supports advisory practice, with a particular emphasis on tax. The cluster reflects the long heritage of the family's enterprise in tax practice, and its skills are correspondingly developed.
The advisory skills in the cluster produce written opinions on legal questions. The opinions follow the structural conventions of formal legal advice, with the issue, the framework, the application, the conclusion, and the supporting authorities each addressed in their own section. The skills are calibrated to the kind of opinion sought, whether a brief informal view or a formal opinion suitable for board consideration or for reliance by counterparties.
The tax skills are the most differentiated part of the cluster. They cover income tax, goods and services tax, customs, transfer pricing, international tax, and the various specialised areas within tax practice. Each is supported by skills calibrated for the relevant statute, the relevant judicial decisions, the relevant notifications and circulars, and the practical conventions that tax practitioners have developed for working with the framework. The tax skills draw, in particular, on the depth of the corpus, which has been developed over decades of careful curation in the family's enterprise.
The tax practice configuration, sometimes called the tax practice pod, combines advisory, tracking, filing, and review skills calibrated for the user's tax practice. The pod operates against the user's particular mix of tax matters, produces the advisory work product that tax practice requires, and supports the ongoing tracking of the regulatory environment that tax practice cannot afford to neglect. The user supervising the pod concentrates on the matters that require senior judgment, while the routine production is handled by the pod under the user's direction.
A specialised cluster of skills supports the work of investigations and dispute analysis, which has its own demands and its own conventions. Investigations work, whether internal investigations conducted by a corporate legal department, regulatory investigations conducted by an authority, or the investigative work that precedes substantial litigation, requires the assembly of evidence, the construction of timelines, the analysis of documentary material, and the production of reports that meet the standards of the audience for whom they are produced.
The skills in this cluster perform document review at scale, extract the relevant facts, organise them into the structures that investigation work requires, and produce the working documents that the investigation team relies on. The chronology skills, mentioned in the litigation cluster, are particularly important here, because investigations often turn on the precise sequence of events. The interview preparation skills support the planning of investigative interviews, with structured outlines drawn from the available material. The report skills produce the formal reports that investigations conclude with, calibrated to the audience and the standards required.
The cluster is often configured as an investigation unit: a team of agents calibrated for the investigation at hand, with the review, extraction, chronology, interview preparation, and reporting skills composed into a coordinated workflow. The unit operates under the close supervision of the senior lawyer leading the investigation, with the routine production handled by the unit and the senior judgment concentrated on the strategic and substantive questions the investigation raises.
Beyond the substantive practice clusters lies a cluster of skills that supports the operational substrate of legal work. Paralegal operations, court operations, document management, deadline tracking, and the many supporting activities that keep a legal practice functioning are each represented in the catalogue. The cluster contains sixteen skills in its current form, calibrated for the operational requirements of contemporary practice.
These skills are, in some respects, the least visible in marketing materials but among the most consequential in daily use. A practice that has automated its deadline tracking, its document management, and its paralegal operations has freed substantial human capacity for substantive work. The skills in this cluster perform the routine activities, with the same discipline of structured outputs and verifiable results that characterises the substantive skills. The integration of these skills into the workflow of the practice is, in our experience, one of the surer paths to broad adoption of the system, because the skills produce visible benefits without requiring the user to change the way substantive work is done.
The cluster also includes a specialised set of skills for the integration of the system with the firm's existing technology environment. Connectors to document management systems, calendaring systems, billing systems, and the various enterprise tools that legal practices operate are supported, with the skills handling the structured exchange of information between the system and the surrounding environment. The integration is configurable rather than imposed: a firm chooses which integrations to enable, with the skills providing the appropriate behaviour in each case.
Above the substantive skills sits a small but architecturally central layer of skills whose function is not to produce legal work but to supervise the production of legal work by other skills. The meta-reasoning layer contains nine skills in its current form. They include skills for citation verification, for cross-validation across multiple authorities, for evidence mapping in which assertions are traced back to their underlying sources, for adversarial testing in which conclusions are challenged from the contrary perspective, and for confidence calibration in which the strength of conclusions is assessed and reported.
The meta-reasoning skills are invoked, in the standard configurations, on the output of every substantive skill that produces legal work product. The output of a drafting skill is checked by a citation verification skill before being delivered. The output of an analysis skill is challenged by an adversarial review skill before being relied on. The output of a research skill is cross-validated against multiple sources before being treated as settled. The meta-reasoning layer is the architectural mechanism by which the substantive skills are made to answer to the standards of legal work, rather than only to the internal standards of the skills themselves.
The skills in the meta-reasoning layer have been described, in their substantive content, in our other writing on safety and on reasoning. We mention them here because their organisational position in the catalogue is worth noting. They are not features added to the substantive skills; they are skills in their own right, deployed alongside the substantive skills in a structured way. The architecture treats them as peers rather than as adjuncts, which makes them visible, configurable, and improvable on their own terms.
The catalogue, taken as an inventory, is the raw material for the system's operation. The system's actual operation, in any deployment, consists of compositions of skills assembled into teams calibrated for the work at hand. The compositions are configurable, as discussed elsewhere in our writing, but several typical configurations recur often enough to be worth naming.
Each of these configurations is a starting point. Firms that adopt the system commonly customise the compositions to their own practice, and the resulting compositions become the firm's standard operating procedures, encoded. The compositions are saved, named, and reused across matters of similar shape. They are, in effect, the firm's institutional memory of how it likes to work, made executable.
The composition layer is the place where the catalogue's diversity becomes actionable. A user who is presented with two hundred and fifteen skills and asked to choose has too many options to navigate. A user who is presented with a configured pod calibrated for the work at hand has, in effect, a team ready to be supervised. The composition layer translates the breadth of the catalogue into the focus of a working assignment.
The catalogue described in this essay is the catalogue of the system at this point in time. It will continue to grow. New skills will be added as new kinds of work emerge, as new jurisdictions are supported, as new user segments are served, and as the architecture is extended. The growth will be deliberate, with each new skill added against the disciplines that have produced the existing catalogue: a real unit of legal work, a defined evaluation standard, integration into the architecture rather than addition alongside it.
The growth will also be opinionated. We will add skills where the work is real, the audience is identifiable, and the standards are reachable. We will not add skills for the sake of the count, or in pursuit of the impression of comprehensiveness. The catalogue's value is in its coherence, not in its number. A coherent catalogue of two hundred and fifteen well-built skills is more useful than a sprawling catalogue of five hundred uneven ones, and we have chosen the former at the cost of the latter.
The catalogue's most important property, however, is not its size or its coverage. It is its disposition to grow in the direction the work demands. The system's users tell us, every day, what the system needs to do better, what gaps they have encountered, and what new work they would like to bring under the system's coverage. The catalogue grows in response to these signals, adding the skills that the users have shown they need rather than the skills the team has imagined. The matrix is, in this sense, alive. It is shaped by the work it is asked to support, in the same way that the corpus is shaped by the questions it is asked to answer. Both will continue to evolve. The architecture is built to absorb the evolution. The discipline that produces the architecture is built to sustain it.
Two hundred and fifteen is a number that, by the time this essay reaches its readers, will already be inaccurate. New skills will have been added. The configurations will have been refined. The matrix will have shifted. What will not have changed is the discipline by which the catalogue is built. It is the discipline that this essay has been concerned with. The number is incidental. The matrix is what we have built. The work it supports is, finally, the point.