14 AI and Compensation

 

Lawyers live by words. When it comes to text, an average lawyer is probably more prolific than someone who writes for a living like a journalist or an author. Lawyers write, read, mark up, litigate and negotiate. Language is at the very core of the profession. While doing so, lawyers have to process and comprehend large amounts of information at the highest level of detail. As this would be undoable, or extremely inefficient and time-consuming for one person, law firms have for long relied on a team of associates and paralegals to research, review and prepare documents. This created the model of leverage which has become the cornerstone of the business of law.

 

Leverage dictates the relationship between labour costs, pricing and profit margins. At its most fundamental level, it is the ratio of associates and non-partner timekeepers to equity partners. By employing a pyramid structure, a law firm can generate revenue that far exceeds the cost of its labour, allowing partners to capture the surplus value created by those working beneath them. The mechanics rely on the billable hour as the primary unit of currency. The goal of a highly leveraged firm is to maximise the spread between what associates are paid and what they bill, across as many associates as possible. This transforms the firm from a simple service provider into a production system where the partners are the owners.

 

In a buoyant market, new technology that increases efficiency is not a threat. Any gain in efficiency was easily offset by growing demand and higher rates. Today the situation is different. The legal market has been largely stagnant in real terms for a number of years, with growth in revenue driven primarily by rate increases. Legal technology and AI will increase efficiency far more profoundly than anything that came before, but this time it cannot be absorbed by growing demand and higher rates. The industry will need to rethink its model. Holding on to time-based billing as AI takes over more of the work will turn technology from friend to foe.

 

 

The first wave: legal technology

The foundations were laid over two decades. Richard Susskind's 2008 book The End of Lawyers? argued that the bespoke, handcrafted way of practising law was coming to an end. IBM Watson's defeat of the human champions on Jeopardy! in 2011 demonstrated that machines could navigate natural language and unstructured information at a level previously assumed to require human cognition. For the legal profession, the implication was direct: the high-volume research and document processing that had been the primary responsibility of junior associates and paralegals was, in principle, automatable.

 

The most tangible early manifestation was ROSS Intelligence, an AI legal research platform built on the Watson foundation. ROSS secured partnerships with major Am Law 100 firms before its closure in 2021 following a legal dispute over its training data[1]. Despite that ending, ROSS and Watson together had broken the psychological barrier within the profession, proving that technology could handle cognitive legal tasks and setting the stage for what followed.

 

 

The second wave

The period between 2015 and 2022 transformed legal technology from a back-office utility into a strategic necessity. The transition from on-premise servers to cloud infrastructure, the emergence of practical AI tools for document review and contract analysis, and a surge in venture capital investment combined to change the technology landscape materially. Cloud adoption in law firms went from below 40% in 2015 to over 70% in most markets by 2022[2], accelerated sharply by the pandemic. Thomson Reuters, through Westlaw Edge and the Practical Law and HighQ platforms, consolidated its position as the primary infrastructure provider. Contract lifecycle management and e-discovery tools entered the mainstream.

 

Then, on 30 November 2022, ChatGPT launched and ended the wait-and-see period definitively. A general-purpose large language model that could draft emails, summarise complex filings and generate initial legal clauses in seconds using natural language was accessible to anyone. The early period produced both excitement and caution in equal measure. High-profile incidents, most notably a US attorney submitting a brief filled with fictitious citations generated by an AI tool, established the hallucination problem as the central challenge for legal adoption. The industry response was rapid: grounded AI solutions that anchored generative models to verified legal databases reduced the risk while preserving the capability.

 

By early 2026 the adoption of AI in law firms had moved past the experimental phase into broad, if uneven, implementation. Large law firms lead with an adoption rate that surveys suggest is approaching 90%, fuelled by their ability to invest in enterprise platforms and dedicated implementation teams. Smaller firms have followed, often through AI features bundled into existing practice management software. The tools that have established themselves as category leaders in legal AI, platforms designed specifically for law firm use with verified legal databases, secure environments and workflow integration, are compressing the time required for document review, research, due diligence and drafting by an order of magnitude compared to human-only work. The question is no longer whether AI will reshape legal practice. It is what the reshaping means for how lawyers are compensated.

 

 

AI and pricing

The billable hour is a mid-twentieth-century innovation, not an ancient tradition. For most of legal history, lawyers charged based on their assessment of the value of the work: annual retainers, fixed fees for routine tasks, value billing at the conclusion of a matter. The shift to systematic time tracking began in legal aid in the early twentieth century and became the dominant model from the 1950s onward, promoted by the American Bar Association as a means of increasing lawyer incomes and adopted widely after fixed minimum fee schedules were struck down as price-fixing in 1975.

 

The adoption of the billable hour fundamentally altered the incentive structure of the profession. It shifted the risk of inefficiency from the law firm to the client: the longer a task took, the more the firm earned. This created a misalignment of incentives that actively penalised efficiency. Culturally, it reduced the lawyer's contribution to a numeric quota. The billable hour became the primary metric for associate bonuses, partnership tracks and firm prestige. The personal cost of this, in burnout and in the narrowing of what the profession valued, has never been fully reckoned with.

 

When I was still a very young lawyer, some of the senior partners used to weigh a file on their hand after completion, to establish the fees for services rendered. Despite being an outdated and disappearing practice at the time, it had an element of fairness: it took into account not only the effort of the lawyers but the value to the client.

 

The Creation-Production Divide© established in Chapter 2 explains why the billable hour has worked as long as it has and why it is now failing. Production work, document review, drafting and execution, generates revenue that effectively subsidises the brief, high-value moments of Creation where a partner's unique judgement makes the decisive difference. AI attacks Production directly. When the cost of Production approaches zero, the subsidy disappears and the only thing that remains commercially defensible is Creation itself. This is not a future scenario. It is what is happening now.

 

The arithmetic is uncomfortable. Consider a firm of 200 lawyers generating revenue of $500,000 per lawyer, with 50 equity partners and a 50% profit margin. The firm invests $1 million in AI that makes its lawyers 5% more efficient. If the amount of work remains the same, the firm is producing the same output in fewer hours. Revenue falls by approximately 5% to $95 million. Costs rise by $1 million to $51 million. Profit falls from $50 million to $44 million: a 12% drop. Profit per partner falls from $1 million to $880,000. The 5% efficiency gain in this example is conservative. Realistically the gain will be higher, and the drop in profitability commensurately larger. This is the erosive effect of increased efficiency combined with a time-based billing model. The firm needs more work to compensate, but the legal market has been growing primarily through rate increases rather than volume. The rate lever cannot be pulled far enough to offset what is coming.

 

In the early stages of AI adoption, more alternative fee arrangements, fixed fees and caps, are to be expected, as both law firms and their clients adjust to a world where hourly billing no longer represents what the work actually cost. The challenge is that partners are generally not well trained in making fixed fees work. It typically starts with a fee proposal that is too low, followed by inadequate management of the execution, compressing the margin further. Getting this right requires new workflows, better data on past mandates, and a discipline of project management that the billable hour model has never demanded.

 

In-house legal departments, meanwhile, are not waiting. Surveys suggest that over 52% of in-house teams have now adopted generative AI, and 64% expect to depend less on outside counsel specifically because of the capabilities they have built internally[3]. The demand side is unbundling into two tracks: for complex, high-stakes matters, elite law firms remain essential, though clients now demand transparency about how those firms use AI to keep hours lean. For everything else, demand is shifting toward fixed fees or moving in-house entirely. Outside the top firms operating in the major legal markets, a substantial portion of current revenue will evaporate in the short term through a combination of greater efficiency per matter and fewer mandates overall, as larger companies do more work themselves. There are only two instruments available to respond: fewer associates, reducing the salary cost base, and alternative fee arrangements that capture the value of AI-enabled efficiency rather than simply billing it away.

 

 

The future of the pyramid model

As AI absorbs the work that occupied the base of the pyramid, the firm structure will evolve from a wide-based pyramid toward a diamond. Fewer junior associates are needed to handle the same volume of work. Those who remain must be more capable from their first day, focused on judgement, analysis and client communication rather than document production. This shifts the associate's role from volume producer to quality optimiser, but it requires a fundamental overhaul of how firms price their services and develop their people.

 

The training implications are significant. The traditional model relied on learning by doing: associates developed judgement by handling large volumes of routine work under supervision. When AI absorbs that routine work, the development path has to be redesigned. Firms will likely move toward simulation-based training and structured exposure to high-level client situations far earlier in an associate's career, accelerating the development that the old model left to time and volume.

 

The compensation implications for the middle of the diamond are equally significant. In the old pyramid, there was a clear path upward or out. In the diamond, the middle must become a destination in itself, not a waiting room. Firms will need to create roles and compensation structures that make senior non-partner positions genuinely attractive as final destinations, with meaningful income, genuine autonomy and a clear professional identity that does not depend on the promise of equity that most will never receive. This is a design challenge that most firms have not yet seriously engaged.

 

 

Compensating the non-lawyer professional

The diamond model does not only reshape the careers of lawyers. It accelerates the arrival of a class of senior professionals who are not lawyers at all, and who will need a compensation framework that law firms have not yet had to design at scale. Chief AI Officers, legal engineers, pricing analysts, knowledge management directors, and senior data scientists are no longer back-office support functions. In the AI-augmented firm, they sit inside client-facing teams, contribute directly to mandate strategy, and in some cases carry revenue and client relationship responsibility of their own. A Chief AI Officer who co-designs a bespoke AI workflow for a major client, presents it in a client meeting, and oversees its execution is performing work that is economically indistinguishable from that of a partner, except that they cannot, under current bar regulations in most jurisdictions, hold equity in the firm.

 

That regulatory constraint is real but should not be treated as permanent. In jurisdictions that have moved furthest toward liberalising legal ownership, most notably England and Wales, where Alternative Business Structures have been permitted since 2012, and Australia, the question of non-lawyer participation has already been partially resolved. The United States and most European jurisdictions have not followed at the same pace. For the foreseeable future, a firm operating in New York or Frankfurt cannot make its Chief AI Officer an equity partner in the formal sense. What it can do is design a compensation structure that functions economically as if it had.

 

Shadow equity is the most practical instrument currently available. Under a shadow equity arrangement, the non-lawyer professional is assigned notional units that track the economic performance of the firm or a defined subset of it, without conferring ownership interest or voting rights. If the firm's profits rise, the shadow equity holder participates in the upside. If profits fall, their variable compensation falls with them, creating the same alignment of interest that equity ownership would produce. Several accounting and consulting firms have used analogous structures, variously called profit participation units, phantom equity or synthetic equity, to compensate senior non-partner professionals for decades. There is no structural reason law firms cannot adopt the same approach.

 

Performance bonuses form the second pillar. Unlike discretionary bonuses for support staff, bonuses for client-facing non-lawyer professionals should be structured against measurable outcomes: revenue generated or enabled, client satisfaction where the professional had direct contact, efficiency gains on mandates they supported, and business development contributions. A senior pricing analyst whose alternative fee proposals consistently convert at a higher rate than the firm average is generating a quantifiable commercial result. A legal engineer whose workflow design reduces a high-volume mandate from forty associate hours to twelve is protecting a fixed-fee margin. These contributions are compensable, and a well-designed bonus structure makes the link between contribution and reward explicit.

 

The governance question is more delicate. A Chief AI Officer who holds responsibility for technology strategy, operational performance, or a significant client relationship has a legitimate interest in the decisions that affect those responsibilities. Excluding them entirely from the partnership forum produces dysfunction. The solution that several firms have begun to adopt is observer or advisory status at partner meetings, with the right to speak on agenda items within their area of responsibility. This falls short of formal voting rights, which in most jurisdictions remain reserved for equity partners, but it creates a meaningful inclusion that respects both the professional's contribution and the regulatory constraint.

 

The broader principle is structural honesty. Firms that recruit senior non-lawyer professionals with partnership-level responsibilities but compensate them as senior associates will not retain them. The talent pool for a Chief AI Officer who can operate credibly in a client-facing legal environment is small and growing more competitive. A compensation model that combines a market-competitive base, a performance bonus tied to measurable outcomes, a shadow equity participation that creates genuine long-term alignment, and a governance role commensurate with operational responsibility is not generosity. It is the minimum required to attract and keep the people on whom the new model depends.

 

 

AI as a team member: the leverage inversion

The discussion so far has treated AI primarily as a process innovation, a tool that compresses the time required for execution, challenges the billable hour, and reshapes the pyramid into a diamond. All of this is true and consequential. But there is a deeper shift underway that has received less attention, and it may matter more than any of the structural changes described above.

 

Think of AI not as software, but as an extraordinarily capable associate sitting in the partner's team. This associate has read everything, every precedent, every deal structure, every regulatory filing relevant to the matter at hand. It never tires, never loses focus, never introduces the variability in quality that comes from asking a second-year to draft a first version at eleven o'clock on a Thursday night. It can run multiple analytical threads simultaneously, synthesise positions across jurisdictions, model scenarios, and produce a first-rate execution pass in the time it would previously have taken a team of four juniors working a full day.

 

Now consider what happens as AI continues to improve. The associate becomes more capable still, better at anticipating objections, better at modelling the other side's position, better at identifying the strategic leverage points in a negotiation before the first meeting takes place. At some point this AI associate becomes not just a capable executor but a genuine strategic co-pilot: able to bring to bear, in real time, the accumulated intelligence of thousands of comparable transactions or disputes, and to present the partner with options and their consequences at a level of depth that no human team could previously provide.

 

But this trajectory has a prerequisite that is rarely discussed, and it is not technical. The partners who will benefit most from AI are not necessarily the most digitally fluent. They are the ones who know how to delegate and how to communicate. Delegation, because even an extraordinarily capable associate produces its best work when the partner gives clear instructions, defines scope precisely, and trusts the output enough to build on it rather than dismantling it. The partner who cannot let go, who rewrites everything and second-guesses every draft, will extract a fraction of the value available to the one who has learned to direct rather than do. Communication, because AI performs in direct proportion to the quality of what it is asked. A vague brief produces a generic result. A partner who can articulate precisely what is needed, what the client’s real concern is, what outcome counts as success, and what constraints apply, is giving the AI associate something it can work with. The partners who have always been clear thinkers and clear communicators will find that AI rewards those qualities in ways the old model never quite did.

 

The implication of this trajectory is the one that the rest of the industry's analysis of AI has largely missed. In the traditional pyramid, leverage was relatively democratic. A partner with twenty associates had leverage. A competent mid-level partner with five associates also had leverage. The ratio of output to personal input was improved for everyone by the same structural mechanism. AI leverage does not work this way. It is not democratic. It is exponential and it is asymmetric in a specific and predictable direction.

 

 

The amplification gap

Consider two partners. The first is technically proficient, reliable, thorough and well-organised. He handles his matters correctly, meets his clients' expectations consistently and generates solid if unremarkable results. The second has something different: a rare intuitive grasp of what the client actually needs as distinct from what they have asked for; genuine creativity in structuring solutions that others would not have found; a deeply calibrated understanding of the human dynamics on the other side of the table, the personal stakes, the political pressures, the unstated constraints that will determine whether a deal closes or collapses. His judgement, in other words, is exceptional.

 

In the old model, both partners were constrained by the same ceiling. Exceptional judgement could only be applied to the matters where that partner was personally present and personally engaged. The hours in the day set the upper limit. The superstar and the solid performer were, in practical terms, producing within the same order of magnitude.

 

AI removes that ceiling, but it does not remove it equally. The solid partner directing AI gets solid work done at greater speed and lower cost. The exceptional partner directing AI gets something categorically different. His strategic insight, his creative instinct, his reading of the human situation: these are now the inputs to a system that can execute on that judgement at a scale and speed that was previously impossible. The AI does not just help him do more; it extends the reach of what he uniquely is able to perceive and decide into every corner of the matter, consistently, without the dilution that previously occurred when his judgement had to be transmitted through layers of associates who understood it imperfectly.

 

The result is an amplification gap between exceptional and average that has no historical precedent in the legal profession. A partner whose judgement is genuinely irreplaceable, who brings something to a mandate that no AI can generate and no other lawyer in the market can match, is about to become capable of producing, with a small AI-augmented team, what previously required a large practice group. The revenue multiple that separates the superstar from the competent is no longer capped by time. It is capped only by the quality of the partner's judgement.

 

Silicon Valley encountered a version of this question earlier and has been working through its answer for the past decade. The concept of the 10x developer, the idea that a single elite engineer could deliver the output of an entire team, originated in a 1968 study of coding performance[4]. In its original form it was a claim about individual speed: the best programmer was simply ten times faster than an average one. What the technology industry discovered through experience was that raw speed was not only the wrong metric but often a destructive one. The engineer who produced massive quantities of code while refusing to document, mentor or collaborate created systems that only they could maintain, bottlenecks that slowed everyone else, and a team culture that drove away the people around them. By 2026 the concept had been substantially revised. The engineer now described as a 10x contributor is not the fastest individual producer but the one whose presence multiplies the capability of everyone around them: the person who makes better decisions that save months of rework, who brings junior colleagues along rather than bypassing them, who builds systems others can extend. The label shifted from individual genius to collective multiplier. The legal profession is arriving at the same distinction now, accelerated by AI. The partner whose value is execution speed is the one AI replaces. The partner whose value is judgement, creativity and the elevation of those around them is the one AI amplifies. The 7-Core Dimensions© are, seen in this light, a description of the multiplier: the qualities that make a lawyer not merely exceptional in isolation but decisive in the context of a team, a client relationship and a firm.

 

 

Which qualities survive and which do not

This is where the 7-Core Dimensions© framework, described in Chapter 4, turns out to be not merely a useful development tool but an accurate map of the post-AI talent landscape. The research showed that 83% of what clients praise in their best lawyers has nothing to do with legal knowledge. It is everything else. In the context of AI amplification, that finding has a new and sharper meaning: the 83% is precisely what AI cannot replicate, and it is precisely what AI will most powerfully extend.

 

Understanding the Business: the ability to see the client's actual commercial situation rather than the legal question as presented is a form of judgement that requires lived experience of how organisations work, how decisions are made under pressure, how a board thinks about risk when a transaction is in its final hours. AI can provide the analysis. It cannot provide the intuition that knows which number in that analysis is the one the client's CFO is actually worried about. A partner who has that intuition, and who now has AI-generated analysis to work with, is more formidable than they have ever been.

 

Creativity: the ability to find solutions that others would not have found becomes more valuable, not less, when AI handles the execution. The creative insight is the scarce input. Before AI, a creative solution still had to be executed by a team that introduced its own friction and variability. Now the execution follows the insight cleanly and immediately. The creative partner is no longer constrained by the gap between what they can conceive and what their team can deliver.

 

Practice Development: the building of relationships, the cultivation of trust, the ability to make a client feel genuinely understood is immune to AI in the deepest sense. Clients do not build relationships with platforms. They build relationships with people who have demonstrated, over time, that they understand what matters and can be trusted under pressure. The partner who has those relationships, and who can now deliver on them through AI-augmented execution, has a stronger competitive position than ever. The partner who was relying on execution quality rather than relationship depth to hold their clients is in a different situation entirely.

 

Practice Management: the discipline of running a matter with clarity, forward planning and consistent communication is the dimension most directly transformed by AI. The partner who has this discipline developed will find that AI makes them dramatically more efficient. The partner who lacks it will find that AI exposes rather than conceals the disorganisation, because the gap between what AI can produce and what actually reaches the client becomes more visible when the execution bottleneck shifts from associates to the partner's own management of the process.

 

People Skills and Emotional Intelligence: the ability to read the room, to understand the unstated drivers on the other side, to know when to press and when to let silence do the work, are perhaps the dimension where the asymmetry is most acute. AI can model scenarios and simulate positions. It cannot feel the shift in a room when a counterpart's confidence breaks, or know intuitively that the person across the table is acting under an instruction they disagree with and can be reached if approached differently. The partner who has this skill, and who is no longer spending their hours on document review, has more of their cognitive energy available to deploy it.

 

Presence and Confidence: the ability to command trust in a room, to hold a position under pressure without becoming aggressive or defensive, to represent the client's interests with an authority that makes the other side take them seriously, is increasingly the primary thing that separates a great lawyer from an AI-assisted one in the client's perception. When both sides of a transaction are using comparable AI tools and the quality of execution is roughly equivalent, what the client is buying is judgement and presence.

 

Integrity: the ability to hold an ethical line when commercial pressure and personal temptation are pushing in the other direction acquires a new dimension in the AI context. The partner who directs AI is directing an instrument of considerable power. The quality of their judgement about when to use that power, and in whose service, and within what limits, is not a soft consideration. It is a central one. The AI does not have a conscience. The partner does. In a world where AI can construct a deeply persuasive case for almost any position, the partner's integrity is the last line between their client being well-served and the profession being corrupted.

 

 

The compensation logic that follows

If this analysis is correct, the implications for compensation are direct and in some ways uncomfortable for firms that have invested in relatively egalitarian models.

 

The traditional argument for compressed compensation spreads, that it is difficult to attribute outcomes to individuals in a team production environment, and that the collaborative culture this compression supports is itself part of what makes the firm valuable, weakens considerably in the AI-augmented model. When an exceptional partner is directing AI associates rather than human ones, the attribution problem largely disappears. The AI creates an audit trail. The strategic decisions are the partner's. The execution follows from them cleanly. The revenue generated is, far more than in the old model, a direct expression of that partner's judgement.

 

This means that partners who sit at the exceptional end of the 7-Core Dimensions© spectrum will, for the first time, be able to demonstrate their contribution with a precision that the old team-production model obscured. They will know their value. Their clients will know their value. And the lateral market, which already moves faster than firms can manage, will know their value and be willing to pay for it. The compensation model that fails to reflect this will not hold these people.

 

At the same time, the partner whose primary value was always execution quality, thorough, reliable, technically sound, but not a standout on the human dimensions, faces a more difficult reckoning. This partner is not without value in the AI-augmented firm: practice management, quality oversight, mentoring the smaller and more capable associate cohort, and the governance of AI-driven workflows are all genuine contributions. But the premium the market will pay for execution quality alone, once AI can deliver execution quality at a fraction of the cost, will compress. The honest conversation about this, in compensation committees, in partner reviews, and in the broader culture of law firms, has barely started.

 

 

The institutional question

There is a question underneath all of this that compensation design cannot answer on its own, though it can make it easier or harder to ask.

 

If the most exceptional partners can, in the AI-augmented model, produce exceptional results with a small team and minimal institutional support, why do they need the firm? This is not a rhetorical question. It is one that the most talented lawyers in every major market are beginning to calculate, and the answer matters enormously for the future of the institution.

 

The honest answer is that the firm still provides a great deal: institutional client relationships that belong to the platform rather than the individual, a brand that clients trust before they have met the partner, the regulatory standing that makes the work possible, the infrastructure of risk management, the complementary capabilities of other practice groups, and the AI platform itself, which at enterprise scale requires investment that individuals cannot make alone. These are real assets. They are the reasons why the most exceptional lawyers in history have, almost without exception, chosen to build within an institution rather than outside one.

 

But they are assets that have to be earned and maintained. The firm that is primarily a cost-sharing vehicle for a collection of individual practices, where the partners share a roof and a billing system but not much else, will find that the new economics of AI removes the last argument for the superstar to stay. The firm that is genuinely greater than the sum of its parts, that provides its exceptional partners with something they could not replicate alone, will find that AI makes it stronger rather than more fragile.

 

This is, in the end, the compensation argument restated at its deepest level. A compensation system cannot manufacture the institutional value that makes a firm worth belonging to. But it can express, in the clearest possible terms, what the firm believes about the relative importance of individual brilliance and collective strength, and whether it has understood that in the AI era, the answer to that question has changed.

 

As the TGO Value Matrix© established in Chapter 2, there is no linear relationship between time and value. The matrix makes clear that what clients are willing to pay depends not on how long the work took but on the return the outcome generates for them and on how many lawyers could have delivered it equally well. In the AI era, that principle governs not just client pricing but partner compensation. The value is in the judgement. Everything else is becoming a commodity.

 

[1] Thomson Reuters Enterprise Centre GmbH v. ROSS Intelligence Inc., US District Court for the District of Delaware, Case No. 20-613 (filed May 2020). Thomson Reuters brought suit alleging copyright infringement in connection with ROSS's use of Westlaw content to train its AI legal research platform. ROSS announced the cessation of its operations in January 2021, citing the litigation as a primary factor.

[2] Thomson Reuters Institute, State of the Legal Market reports (2016–2023); International Legal Technology Association (ILTA) Technology Survey. Cloud adoption figures for law firms reflect data compiled across these annual surveys. The pandemic-driven acceleration is documented in the Thomson Reuters Institute 2021 and 2022 reports and in the Wolters Kluwer Future Ready Lawyer Survey 2022.

[3] Association of Corporate Counsel (ACC), Chief Legal Officer Survey 2024/2025; Bloomberg Law, "Legal Operations Technology Survey," 2025.

[4] Harold Sackman, W.J. Erikson and E.E. Grant, "Exploratory Experimental Studies Comparing Online and Offline Programming Performance," Communications of the ACM, vol. 11, no. 1 (January 1968), pp. 3–11. The study documented a 10:1 to 25:1 performance range between the strongest and weakest programmers in experimental tasks, giving rise to the "10x developer" concept  - though the original paper's methodology has since attracted critical scrutiny.