Almost every professional service firm on the planet operates on the fundamental assumption that services are created and delivered to clients by humans. In legal services, this assumption is codified in the regulations governing the practice of law. Except at the most routine levels of service, clients expect the services rendered to them to, at the very least, be fronted by human lawyers. So what does the term AI-native law firm even mean?
The word native has a precise meaning. In the context of technology, for instance, saying that a firm is cloud-native does not simply mean that it uses cloud computing. It means that the firm was designed — its strategy and its fundamental business processes — around cloud computing. A firm that moves its servers to the cloud is simply using a new tool. A firm built from the start around what the cloud makes possible is entirely different. The first bolts a new capability onto an old structure. The second is shaped by the capability from the outset.
Similarly, an AI-native law firm is not one that has simply acquired AI tools and is encouraging its people to use them. It is a firm whose ways of working — its client value propositions (CVPs), its resources and the ways in which it is organised — its fundamental recipe for sustainable competitive advantage — are built upon the assumption that the bulk of the work is done by AI.
Note: In the paper, I use the four components of CVP that I usually prefer: substantive solution; client experience (CX); cost; trust. I find these the most useful in thinking about how the value propositions that clients expect from their law firms, and how those firms respond, are evolving. Along with Resources and Organisation, CVP is one of the three primary pillars for my model of what drives performance and [should] drive strategy.
In 2026, agentic AI systems have very suddenly become a reality. As these advance, lawyers and other humans will find themselves doing quite different things to what they did in conventional firms until even quite recently. They will spend their time defining the context and substance of client legal needs with enough precision to map CVPs to these in ways that clients find compelling enough to select them over rivals. Similarly, they will have to define precise outcomes that must be achieved and standards that must be met. They will apply their professional judgment to ensure that what reaches the client meets their needs. The production work, though – the drafting – will be handled almost entirely by agentic AI systems.
Very few law firms think about AI this way. It is probably fair to say that regulatory and other constraints prevent AI-native law firms from even operating to their full potential today. For most lawyers, AI is seen as a tool to help them work incrementally “cheaper, better, faster” within current workflows. What happens though when these workflows become redundant and, more so, when the new ones that replace them have lifespans of no more than a year or two? It is here that AI-native firms – mostly new entrants to the market – might have an edge.
Almost all law firm leaders know that the way in which their firms operate will change but what that means in practical terms is vague. The future is opaque. We know that AI is advancing at a prodigious rate. What applies today might, in mere months, be quite different. Almost all organisations — clients, law firms, courts, regulators and others — are battling to even begin to understand the scale and scope of the change tsunami that approaches. To add to that, as Clayton Christensen describes in his Theory of Disruptive Innovation, organisational inertia is a powerful handbrake to change. Even when the path forward is clear, that path can be very hard for a firm to take. Yet the rapidly widening gap between new capabilities that AI offers to enhance CVPs and performance and what most law firms actually offer creates powerful opportunity for firms that can bridge that gap, even partially.
A framework for thinking about this
Consider a framework of six levels of AI adoption, ranging from Level 0 (law firms that do not use AI at all) to Level 5 (a theoretical end-state where the production of legal work is handled entirely by AI). Consider these from the perspective of a law firm leader whose own experience might sit at Level 0 or Level 1 and for whom the practical reality of Levels 3, 4 and 5 may be difficult to even visualise.
At the core of the evolution lies how the firm’s lawyers interact with AI. How others in the firm (e.g. business services) do that is obviously also important, but as one progresses to Level 3 and above, so the impact of AI on the way that the firm’s substantive services are created and delivered to clients becomes fundamentally intertwined with the skills that lawyers need in order to work properly. The gap blurs between what lawyers must do themselves and what can be provided as a service by the firm’s IT department or an arm’s-length contractor.
Different practice areas lend themselves differently to AI so a single firm might display different levels across its organisation. This is normal. Areas that lend themselves more to AI can be useful test beds for disciplines for which other practices are not yet ready. However, it would be a mistake to relegate this to fringe practices that are not core to the firm’s business, or areas that are underperforming for systemic reasons that cannot be remediated with AI.
Four Prompting Disciplines
Each level narrative is structured around a framework for understanding the four distinct disciplines that now constitute prompting. With the emergence of agentic AI, the word prompt has a different meaning today (mid-March 2026) to even six months ago. Four fundamentally different skills have diverged as prompting has moved beyond being simply an iterative, vocabulary-based activity performed with LLMs. This shift is profound. So too are the mindset shifts needed for lawyers to properly adapt – to build and deliver compelling CVPs – and the resources needed to support that, and the organisation required to optimise performance.
To a greater or lesser degree, depending on circumstances, all four prompting disciplines are already well in play. The sequence presented in the levels correlates with the way in which lawyers interact with AI, and hence the sources of sustainable competitive advantage and the drivers of performance, too.
- Prompt Craft: the original skill of writing clear, well-structured instructions in a conversational session and iterating on the output in real time.
- Context Engineering: curating the entire information environment within which the AI exists and operates. That is not just the prompt, but the surrounding system instructions, the firm’s knowledge databases, the client’s business and legal context, market information history, firm conventions and quality standards that shape output quality.
- Intent Engineering: encoding organisational purpose, values and decision hierarchies so that the AI system understands the outcome that the client expects, what the firm wants to achieve, and how to resolve competing priorities.
- Specification Engineering: writing complete, structured descriptions of the required output that are precise enough for an autonomous AI system to execute against over extended periods without human intervention. This includes scope, standards, constraints, acceptable trade-offs and acceptance criteria.
Agentic systems can now execute highly complex tasks in one shot, taking minutes, hours or even days or more before producing final work product. It follows that initial instructions provided must include all information the system requires to do that properly. The essential skill required is not dissimilar to instructing a junior to perform a task, but what is different is the AI producing a final product rather than a first draft for a more senior lawyer to redline and iterate.
At all levels, one must differentiate between synchronous and asynchronous interaction. At lower levels, lawyers sit in front of a chat window and iterate – they type, read output, correct and refine. They are the live quality-control loop. At higher levels, the AI works autonomously. There is no opportunity to iterate once the system is running. The quality of the output becomes a direct function of the quality of the upfront definition. A far higher quality of lawyering is required to be able to do this kind of prompting properly. Given its novelty, very few lawyers (or other humans) have the skill yet to do this properly. Few if any law firms have their information organised in ways that are sufficiently machine-accessible and clear and unambiguous to ensure precision.
Level 0: Start line
At level 0, there is no interaction between lawyers and AI. The firm has either not discussed AI at all or such discussion has been limited to prohibiting its use. Junior lawyers almost certainly use general-purpose tools on their personal devices, but without supervision, quality controls, or disclosure. Partners either do not know this is happening, or turn a ‘blind eye.’ The firm’s official position is that legal work is done the way it has always been done: by lawyers, with conventional tools. Clients are assured that quality remains excellent and warned of hallucinations and other risks.
Because no deliberate interaction with AI occurs within the firm’s sanctioned workflows, no prompting discipline applies at this level. But the absence of any discipline is a major risk. Whatever ‘ad hoc’ use is occurring would be classified as rudimentary Prompt Craft at best. That is, unstructured questions being typed into a free or cheap chat tool, with no firm context, no quality standards, no client-specific parameters, no assurance of client privilege or other confidentiality, and no systematic evaluation of whether the output is reliable. The risk is not that lawyers are using AI badly. It is that the firm has no visibility into how AI is being used at all, and therefore no capacity to manage the risks or capture the value. The firm’s data is organised to optimise human, not machine access.
The CVP
The firm’s CVPs are based on traditional capabilities, relationships and reputation. These might seem unaffected in the short term but are eroding invisibly. Substantive solutions delivered to clients remain constrained by the production capacity of humans working without augmentation; Cost structures are static, while competitors begin to shift theirs. Client Experience (CX) is the same as before but clients whose other advisers are moving faster will soon begin to notice. Trust is the most immediate risk: undisclosed, uncontrolled AI use by junior staff creates exposure that the firm’s leaders cannot see and therefore cannot manage. As to resources, the firm remains reliant on those that have traditionally proved successful. Similarly, the way in which the firm is organised. But an iceberg looms ahead.
Level 1: Baby steps
At Level 1, lawyers have subscriptions to approved SaaS AI tools and the firm is encouraging them to use them. They use these tools the way they might use a capable but unfamiliar research assistant: guardedly, for contained tasks, checking everything that comes back. They might for instance ask an AI tool to summarise a lengthy shareholders’ agreement. They might paste in a set of board minutes and ask for a chronological timeline of key resolutions. They might type rough paragraphs of advice and ask the tool to improve the prose. They might ask it to help organise their emails. Each interaction is a self-contained exchange: type a request; read the output; decide whether it is good enough; use it if so or iterate/discard if not.
They learn through trial and error that more specific instructions produce better results. They discover for instance that telling the AI the jurisdiction, the type of client, and the intended audience materially influences the quality of what comes back. They begin to develop a sense for what the tool does well (e.g. summarisation, structural organisation, first-draft prose) and where it is challenged (e.g. novel legal reasoning, nuanced risk assessment, anything requiring knowledge of the client’s commercial context.) But each session starts from scratch. There is no accumulated context. The AI tool knows nothing about any particular client, matter, or the firm’s standards and know-how. Every interaction requires the lawyers to re-establish the basics. They tend to avoid this because the effort required feels disproportionate, especially for quick tasks.
The critical limitation at this level (and the ceiling of Prompt Craft) is that the lawyers are the entire quality layer. They read every word. They catch mistakes in real time. They supply missing context when the output drifts. This works for small tasks but does not scale. It changes little to nothing about how the firm operates or the value that it offers its clients.
The CVP
In a Level 1 law firm, substantive solutions are marginally improved over Level 0. Routine tasks are completed more quickly, freeing modest amounts of time for higher-value thinking. CX is largely unchanged because clients sees no difference in what is delivered, or how. Cost impact is real but limited to individual efficiency gains that are difficult to aggregate across the firm. Provided that human checking is adequate, trust is not yet under pressure.
The fundamental constraint is that the firm has simply attached new capabilities to its old operating model. Changes to the firm’s key resources are minimal, except that junior hiring slows as senior lawyers discover that AI can take on work that traditionally done by juniors. Concerns arise about talent pipeline. Digital capital takes the form of SaaS subscriptions, not integrated assets. Structural capital (the firm’s organisational structures, processes, systems and workflows) remains untouched, as does the way in which the firm is organised.
Level 2: Walking then running
At Level 2, lawyers progress from simple Prompt Craft to the next discipline: Context Engineering. The quality of the output now depends not just on how well lawyers phrase their request, but on the quality and accessibility of the information that surrounds that request in the AI’s operating environment. They have learned, for instance, that the AI produces materially better analysis when it has access to the relevant contracts alongside the correspondence, when it is told what law applies, and when it understands the procedural stage of the matter. They spend time curating this context—selecting the right documents to include, providing background on the dispute, specifying the standard of analysis required—because they have discovered that an under-contextualised request produces superficially impressive but substantively unreliable work. The firm has begun the task of codifying its internal information to enhance access by machines and improve quality (sharpening clarity, resolving conflicts and ambiguities, etc.) Early experimentation emerges with agentic AI, in low-risk areas.
For instance: a disputes team is preparing for a commercial arbitration. The matter involves tens of thousands of pages of different documents: correspondence, contracts, board papers, expert reports and witness statements. The lawyers use the firm’s AI platform (not a general-purpose chat tool, but a system connected to the firm’s document management environment) to run a first-pass review of the entire document set. They ask it to identify every reference to a specific contractual term across all documents, to flag inconsistencies in the factual chronology, and to produce a structured research memorandum on a point of law that arises from the evidence.
The lawyers still read and check everything. The volume of AI output has increased dramatically, so the checking burden has increased with it. Lawyers now spend more of their day reviewing AI-generated work than producing work from scratch. The nature of the errors changes, too. Obvious, outright errors are less frequent. More common are subtle mischaracterisations such as a case cited for a proposition it does not quite support, a contractual interpretation that is plausible but misses a critical qualifier. These flaws are harder to catch. They demand the kind of deep subject-matter knowledge that only experienced lawyers possess. Proper use of AI does not destroy demand for deeply experienced lawyers. It increases the need for them. But is also requires that they reinvent themselves around the new skills sets required.
Most if not all law firms who consider themselves AI-native are likely only Level 2. While they believe they have transformed their practice, they have simply shifted the lawyer’s role from producer to reviewer. This is a meaningful change but not a fundamental one. Lawyers remain the sole quality layer. The constraint has moved from “how fast can we draft this” to “how carefully can we check what the AI drafted.” Checking remains synchronous.
The CVP
Ironically, while many in Level 2 firms “feel” that they are benefitting from AI, the hard reality is usually that it does not yet reflect in real cost saving. Clients, especially those observing AI-induced savings in their own organisations, express frustration at law firms not passing savings on but those savings are taken up by checking taking more time and subtle tensions induced by corporate inertia and investment in new resources to support the AI transformation.
Quality of substantive solutions does improve meaningfully as a firm progresses to Level 2. Firms find they can handle larger volumes, conduct broader and more complex searches, and identify patterns across document sets that would have been impractical using manual analysis. CX also shifts, for the better. For instance: clients benefit from faster turnarounds and more bespoke, timely communication. Tensions deepen within the firm’s economic model, though. Tasks that once required junior associates working for eight hours now requires two hours of AI processing and one hour of partner review. Clients see these new economics and become more strident about passing on savings. The firm’s operating model must begin to address efficiency alongside excellence. Pricing models must become more sophisticated. Governance becomes more complex as AI-generated work products require more thorough quality-assurance processes. Trust can come under pressure if client concerns arise about quality and confidentiality. Firm debates must decide how much to disclose about AI’s role, and how to assure clients that the checking is rigorous enough, and how to mitigate when things go wrong.
Level 3: Managing machines
In Level 3 law firms, lawyers direct AI and judge its output rather than using it to help them do the work themselves. This is a critical inflection point. A lawyer using AI to help with their work is very different to one directing the AI to do the work. The difference might seem subtle, but it is profound. The software equivalent would be a programmer checking the accuracy and functionality of the software produced, rather than the code itself. That AI-generated code may even be unintelligible to the programmer, rendering line-by-line impossible.
This shift means that the prompting discipline must also shift up a gear, from context engineering to the third discipline: intent engineering. These disciplines are sequential. Intent engineering requires precise context engineering. It ensures that the AI understands what the firm and its client are trying to achieve and, critically, how to resolve competing priorities. At the same time, improvements in the first two disciplines allow the firm to take on more complex work than was possible before, at greater scale.
For instance: A team of senior associates in a real estate finance practice no longer draft facility agreements. Instead, they direct agentic AI tools to draft them. Their working day is now reorganised around very different activities. They define the parameters of each document, instruct the AI system with sufficient precision to produce a usable draft, and then evaluate the output against their professional judgment of what the client needs and what the counterparty will accept. Should the draft favour speed of execution or comprehensiveness of protection?
The thinking they do before they engage the AI has become the most consequential part of their work. They must specify: the nature of the transaction and its commercial rationale; the identity and character of the borrower; the lender’s risk appetite and known policy positions; the jurisdictional and regulatory context; the precedents the firm would normally use and the extent to which this transaction departs from them; the points that are likely to be negotiated and the positions the client is likely to take. If they omit any of these (say, simply instruct the AI to draft a facility agreement for a £50m term loan) the output will be technically competent but commercially naive. It will contain none of the bespoke provisions that make the work product useful to a particular client in a particular deal under particular circumstances.
Where the borrower’s commercial needs conflict with the lender’s standard risk controls, which takes priority in this particular negotiation? These are judgment calls that senior associates absorb over years of practice. The AI needs them stated explicitly because it cannot infer them from experience it does not have. And because agentic AI typically executes tasks not iteratively but in a single shot, all that needs to be included in the prompt – up front
The key constraint is no longer how fast the work gets done but how well the work is defined. This governs quality. Flawed instruction to a junior lawyer is retrievable because the junior can walk back down the corridor and asks for clarification. Errors in early drafts are corrected. A vague instruction to an AI system produces a wrong output, confidently and perhaps even convincingly presented, that the lawyers must then diagnose and correct. Especially in early days, this might take longer than producing the work from scratch, with clients reluctant to pay for the extra work. The ability to prompt in precise, comprehensive, terms – understanding the client’s context and commercial objectives deeply enough to articulate them unambiguously before any drafting begins – becomes the defining professional skill.
It is at this point that the answer to the question of how juniors should be trained becomes clear. In order for any lawyer to be able to prompt as required at Level 3 or above, they must be not only well enough versed in the law and the client’s context but also able to express things in terms that are clear, accurate and unambiguous enough for the AI agent to do its work properly. If anything, this implies a far more sophisticated level of skill than that required at Level 0 or 1. It follows that training of these skills needs to start during a lawyers first year in practice or, better still, at law school.
The CVP
It is at Level 3 that a firm’s operating model begins to fundamentally restructure. Capacity is no longer constrained by human production limits. A small team of experienced lawyers directing agentic AI systems can produce work that previously required many more people. CX also transforms if (a crucial condition) firms invest the upfront effort to understand the client’s context deeply enough to specify it accurately. Otherwise, the output risks misalignment with client expectations. Law firms that excel at Level 3 will be those whose lawyers are trained and incentivised to spend time on deep client understanding (far beyond what we today call “client listening”) rather than billable hours.
Law firm economics evolve radically during Level 3. The leverage model shifts from associate hours to AI-augmented senior judgment. The most important resources – those most closely associated with sustained competitive advantage – also change quite radically. Head count becomes far less important; skill to operate in the Level 3 AI context far more so. Digital capital becomes a core strategic asset rather than an operational tool. Structural capita – the firms systems and structures and processes – becomes the enabler for context and intent engineering at scale. Trust becomes the critical differentiator: clients must trust that the lawyers’ judgment, applied to AI-generated work, meets or exceeds the quality of conventionally produced work. Law firms that perform well at Level 3 can expect revenue per lawyer (RPL) to multiply as market share expands.
Level 4: Pushing limits
At Level 4, lawyers no longer review the AI’s actual work at all. They review its conclusions and other output as finished product. This is made possible by the fourth dimension of prompting: specification engineering. That means, to write a detailed specification for how the work is to be executed, including the defining of scope, acceptance criteria, jurisdictional parameters, risk tolerances, and client-specific considerations. The lawyers then step away. When they return, they assess whether the output meets the specification. They care whether that output works, not how it was assembled.
This might sound appallingly risky, but the notion of judging AI on its output rather than its content is already well with us. Even highly experienced coders find routinely that when they “vibe-code” something (i.e. prompt an AI to build it by coding it itself, instead of them writing that code themselves) the code produced is unintelligible to them. But the output works perfectly. This principle might conflict with the fundamental requirement that legal work must be explainable, but it is a reality of today’s world. It is the quality of the lawyer’s specifications (and also intent and context engineering, and general prompt craft) that ensures that clients are properly advised and represented, and avoids malpractice claims.
For example: Following a change in sanctions legislation, a team of partners leading a regulatory practice have been asked by a multinational client to review its compliance framework across fourteen jurisdictions. Under conventional methods, this would require lawyers in each jurisdiction, months of elapsed time, and overheads that would consume a significant portion of the budget. The partners do not assemble that team. They write a specification.
A specification might be considered an equivalent of detailed instructions to counsel. It defines the scope of the review: which entities, which regulatory regimes, which categories of transaction. It sets the standards the analysis must meet: the level of legal certainty required, the treatment of ambiguity, the extent to which local regulatory guidance (as distinct from primary legislation) should be taken into account. It articulates the client’s risk appetite (this client is conservative on sanctions compliance and has communicated explicitly that it would rather over-comply than face enforcement risk.) It specifies the output format: a jurisdiction-by-jurisdiction matrix, a consolidated risk register, and a prioritised action plan. It defines the steps that must be followed in executing the work. And it defines what “done” looks like: specific acceptance criteria against which the completed work can be evaluated without the partners having reviewed the underlying reasoning in each jurisdiction.
The partners have written a document precise enough for an autonomous AI system to execute against across multiple jurisdictions, over an extended period, without monitoring at each step. They do not review how the AI reached its conclusions in each jurisdiction. They review whether the conclusions meet the specification. They care not how the work was performed, but whether the output works to specification.
The quality of this specification depends deeply on how well the partners understand three things: the law, the client, and the client’s commercial context. Agentic AI does not ask for clarification. It builds exactly what is described to it. If the specification is ambiguous about how to treat a jurisdiction where sanctions guidance is evolving but no formal legislation has been enacted, the AI will make a choice. That will be based on statistical plausibility rather than professional judgment – and might easily be wrong. To avoid such errors, the partners must anticipate that question, and every other, and answer these in the specification before the Agentic system began its work. No opportunity exists to iterate. The requirement for deep client understanding (which has always been central to excellent legal practice) becomes absolute.
As to training juniors, the points made under Level 3 become even more pressing. The skills required to practice in a Level 4 law firm are arguably more sophisticated and wide ranging than those in a Level 0 – 3 firm. The margins for error are tighter. The work delivered is far more complex, at greater scale and more otherwise challenging than is possible in Level 0 – 3 firms.
The CVP
Level 4 transforms every dimension of the firm’s operating model. Solutions that were uneconomic in lower levels become practical. Costs reduce to a fraction of what applies under earlier levels, so too time required. Processes that drive high CX are also heavily automated. Cost collapses for the production component of the work while the value of the specification ramps up. Cost and quality advantage combined with exponentially greater capacity allows the firm to hoover up market share, driving profit to multiples of what the top law firms generate today. If RPL multiplying is not achieved in Level 3, it certainly is in Level 4. Most Level 0 – 2 firms are unlikely still to survive.
The firm no longer sells effort measured in hours. It sells depth and precision of its collective professional judgment, encoded in specifications that direct the AI. Trust is a major limiting factor, though. No regulator has yet given clear guidance on how far lawyers can rely on AI-produced work without personally reviewing the underlying reasoning. Until that guidance arrives, the firm carries a risk it cannot easily quantify.
Level 5: Dark chambers
A hypothetical agentic AI system that turns legal specifications into finished work product. No lawyer drafts. No lawyer reviews drafts. Specification in, client-ready output out. In the manufacturing world, this is called “dark factory”- a reference to factories that operate with the lights off because no human is present on the production floor. A handful of software engineering teams might operate there today. Obviously, no law firms are. But it is worth understanding as a destination, hypothetical or otherwise, because the trajectory of the technology points unambiguously in this direction.
Imagine a managing committee presiding over a firm that has been redesigned around production of legal work being handled entirely by agentic AI systems. No lawyer drafts. No lawyer reviews drafts. Specifications go in and completed work comes out. Five years ago, last year, or even today, such a premise stretches the bounds of imagination. But logic dictates that this Level might well emerge within the foreseeable future.
The lawyers in this firm do not interact with AI through prompts, or even through the kind of specifications described at Level 4. They operate through an institutional architecture – a structured, continuously refined system of client knowledge, matter templates, quality standards, regulatory constraints and decision hierarchies that the firm has built over time and that governs how AI systems produce work. The individual lawyer’s contribution is not a specification for a single piece of work. It is the ongoing refinement of the institutional specification. This means deepening the firm’s understanding of each client’s commercial context, risk appetite and strategic priorities. It means updating the decision frameworks as regulations evolve. It means designing the evaluation criteria against which completed work is assessed.
All four prompting disciplines are present, but they have been absorbed into the firm’s infrastructure. Prompt craft is embedded in the templates. Context engineering is embedded in the knowledge management systems. Intent engineering is embedded in the client relationship architecture: structured records of each client’s values, priorities and decision boundaries that are maintained as living documents. Specification engineering is effectively the operating system of the firm. It is the means by which human judgment is translated into machine-executable instructions at scale.
In a Level 5 firm, it is a long time since humans were the limiting factor. It is no longer even the technology that limits. It is the quality of the institutional knowledge. The firm that operates at Level 5 has not eliminated the need for lawyers. It has elevated what lawyers do to at the highest possible level of abstraction. That is, understanding what the client needs at least as well as the client does in contexts far more complex than exist today. It means defining precisely and comprehensively the standards that work must achieve, to meet those needs. It means exercising the professional judgment that determines whether the finished product serves the client’s interests as well as is possible. It creates the opportunity to be what David Maister called a Trusted Advisor, but at a far greater degree of scale and complexity than was ever possible before. Everything else is production. And production is what the machines do.
The CVP
At Level 5, substantive solutions are limited only by the depth of the firm’s institutional understanding of the law and its clients – not by production capacity, which is effectively infinite. CX is tailored in exquisite detail to the expectations of each individual client. Cost approaches the marginal cost of computation for the production component. The firm’s pricing reflects the value of its judgment, not hours consumed in delivery. Trust becomes the very bedrock of the firm’s CVPs.
Conclusion
The admonition that it is dangerous to make forecasts, especially about the future, is variously attributed to Niels Bohr, Samuel Goldwyn, Yogi Berra, Mark Twain and Nostradamus (amongst others.) This is not the intent of this paper. Rather, it is to apply logic observed in business generally and especially in the development of agentic AI, to model the key themes that will unfold as its capabilities advance. Because advance it will – of that we can be sure. It is intended to provoke thought amongst leaders of firms at Levels 0, 1 and 2 about how their sources of sustainable competitive advantage and their firms’ performance drivers are evolving.
It is intended to bring into stark relief how different a law firm’s business model will be when advances in AI drive client needs to the point that Levels 3 and 4 emerge, and how deeply uncompetitive their current business models will be with those Level 3 and Level 4 firms. We can leave the possibility of Level 5 firms as purely speculative, at least for now.
Organisational inertia is a powerful drag but the gap between increasingly complex AI-induced client needs and increasingly capable tools to meet those on the one hand and, on the other, the quickly declining ability of the more Luddite law firms to do that is creating unprecedented opportunity for firms that are willing and able to recognise and seize that opportunity.
It will be a rocky ride, but far more so for firms that fall behind than those who pull ahead.
Those who like me follow Nate Jones’ work will recognise how he inspired this paper.
What do you think?
