Purpose, Execution, Judgement: the paper behind the framework

Rob Dixon

Mar 10

7 min read

Dixon, R.J., 2026b. Purpose, Execution, Judgement: A Framework for Human-AI Collaboration and Organisational AI Capability in the Post-ChatGPT Era. SSRN Working Paper, 31 pages, posted 10 March 2026. Available at dx.doi.org/10.2139/ssrn.6282200.

In brief. The first of what I intend to be a series of SSRN papers. The claim in one sentence: every productive human-AI activity has three domains, Purpose (why), Execution (how), and Judgement (which output matters), and an organisation only gets value from that collaboration when five multiplicative capability dimensions are in place underneath it.

Why I wrote it

I first set out Purpose, Execution, Judgement in an article last summer, prompted by a question that came up in nearly every workshop I ran: where does the human stop and the AI begin? (Dixon, 2025a). The capability side of the argument is older still; I described the five dimensions of organisational AI capability in 2024 (Dixon, 2024c), and both frameworks run through the Playbook (Dixon, 2025d).

What neither article did was position the ideas against the research literature, join them into one model, or state what they predict in a form someone could test. That's what a working paper is for. Consultancy and teaching are legitimate places for a framework to come from, and the paper says so plainly, but it has to earn its place alongside the human-AI teaming literature rather than sit beside it. So this is the academic version.

What it argues

Organisations scrambled to adopt generative AI after November 2022 without a coherent way of deciding where human responsibility ends and machine capability begins. The result shows up as prompt drift at the individual level, automating the wrong activities at the team level, and enterprise models that never reach production. The literature I reviewed is good on productivity effects and on taxonomies of interaction, but it under-theorises judgement, treating it as a check on machine output rather than a distinct value-creating activity, and nothing addresses at once the micro division of labour, the organisational conditions that support it, and how the two connect. The paper tries to fill that gap with two frameworks and their integration.

The first is PEJ. Purpose, Execution, Judgement (PEJ) is a framework for human-AI collaboration in which every productive activity, regardless of scale, comprises three interlocking domains: Purpose, the human-defined why; Execution, the collaborative how; and Judgement, the human-applied decision about which output matters. Purpose stays human because it's where direction and intent live, and it's no longer a once-a-year strategy exercise but a continuous activity. Execution is a partnership, and the paper models its effectiveness as Talent × AI Literacy × AI Power. Judgement stays human for three reasons: it needs contextual understanding beyond any model's training distribution, it carries consequences, and without it you get what Simon Willison named AI slop (Willison, 2024), output that's abundant, cheap, and valueless. When AI makes execution abundant, value migrates from production to selection. An AI can draft fifty versions of a contract before lunch; the human who decides which one ships is doing the highest-value work in the chain.

The paper's recurring image is the film director. Scorsese said he wanted Goodfellas to "begin like a gunshot and have it get faster from there". That was his purpose. Feature films typically shoot between 10 and 95 hours of material for every hour of finished screen time (Morrow, 2016), so the film is not what was shot; it's what he chose. AI can produce the footage. Only the human decides what the audience sees. Before November 2022 most knowledge workers were both director and crew. The shift is from worker to director.

The second framework is AI Capability: Organisational AI Capability = Talent × Purpose × AI Literacy × AI Tools × Data Infrastructure. The multiplication is the whole point. In an additive model, strength in one dimension compensates for weakness in another. In a multiplicative model, a zero anywhere renders the product null. Skilled people and good tools with no coherent purpose produce activity that isn't aimed at anything; clear purpose and strong data with no AI literacy can't turn those assets into outcomes; tools without data produce generic output.

The integration is the part that's new. AI Capability answers what an organisation needs; PEJ answers how humans and AI collaborate once it's there. Together they form a two-layer architecture: a capability layer that establishes the substrate, and a collaboration layer that governs the workflow. Organisational purpose, a readiness condition, nests within and enables individual purpose, a workflow activity, so weak purpose at the top degrades purpose-setting at the desk even when talent and literacy are strong. When an initiative fails, leaders can ask which layer broke. In my experience the answer is usually both, which is why you need the integrated model.

The evidence, and what it predicts

The paper leans on two field studies. Brynjolfsson and colleagues found a 14 per cent average productivity gain when 5,179 customer support agents got an AI assistant, rising to 34 per cent for novices and minimal for experienced staff (Brynjolfsson, Li and Raymond, 2023). Dell'Acqua and colleagues found consultants using GPT-4 were faster and better inside the model's capability frontier and worse outside it, the jagged frontier (Dell'Acqua et al., 2023). Both fit a model in which literacy and talent interact rather than add, with the task deciding which one is binding. Hemmer and colleagues' distinction between complementarity potential and complementarity exploitation makes the same point from the other side: capable AI is necessary but not sufficient, you also have to design the roles and processes around it (Hemmer et al., 2025).

Six propositions follow: multiplicative capability; purpose primacy (initiatives with explicit, continuously reviewed human purpose outperform those where purpose is vague or delegated); judgement as value differentiator; a literacy-talent interaction that varies by task type; level agnosticism (PEJ holds at individual, team, and organisational levels, and failures can be traced to imbalance at one or more); and integration necessity. All six are stated so they can be falsified, and the limitations section is clear that they haven't been yet.

What it means in practice

Treat PEJ coverage as a non-negotiable design constraint when you form a team. The question isn't whether one person can master all three domains but whether the team does: absent purpose leads to automating yesterday's processes, absent execution produces slideware instead of software, absent judgement produces a flood of slop. Use the capability formula as a diagnostic: audit each dimension, find the weakest, and put your money there, because in a multiplicative system that's where the marginal return is highest. Redesign performance management to measure purpose, execution, and judgement as distinct competencies, since volume metrics become perverse when AI can inflate volume at will. And ride the ratio. The human share of execution is on a one-way trajectory towards machines, and the gap between an AI-literate professional and one who isn't doesn't grow linearly. It compounds.

If you've followed the talks, you'll recognise the argument: it's the "AI-literate knowledge worker" case I made at LTSE in 2023 (see AI won't take your job) and the strategy line I drew for recruitment leaders in 2024 (see Our Viking moment), now with the scaffolding underneath. The paper points to a companion piece on the ethical and economic side: accountability, who captures the value, and the shift from a knowledge-based to a purpose-based economy. I've set that out as an article (Dixon, 2026a) and I'm working it up as the next paper.

How to cite it

Dixon, R.J., 2026b. Purpose, Execution, Judgement: A Framework for Human-AI Collaboration and Organisational AI Capability in the Post-ChatGPT Era. SSRN Working Paper. Available at: https://dx.doi.org/10.2139/ssrn.6282200

References

Brynjolfsson, E., Li, D. and Raymond, L.R., 2023. Generative AI at Work. NBER Working Paper 31161, April. Available at: https://www.nber.org/papers/w31161 [Accessed 4 September 2026].

Dell'Acqua, F., McFowland, E., Mollick, E.R., Lifshitz-Assaf, H., Kellogg, K., Rajendran, S., Krayer, L., Candelon, F. and Lakhani, K.R., 2023. Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality. Harvard Business School Working Paper 24-013, 15 September. Available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4573321 [Accessed 4 September 2026].

Dixon, R.J., 2023. AI won't take your job. The AI-literate knowledge worker will. [online] robdixon.ai, 25 May. Available at: https://www.robdixon.ai/post/ai-literate-knowledge-worker-ltse2023 [Accessed 4 September 2026].

Dixon, R.J., 2024. Our Viking moment: why the winning strategy changed on 30 November 2022. [online] robdixon.ai, 4 September. Available at: https://www.robdixon.ai/post/our-viking-moment-strategy-change-ai [Accessed 4 September 2026].

Dixon, R.J., 2024c. Defining AI Capability: A Holistic Framework for Organisational Readiness in an AI-Driven World. [online] LinkedIn, 18 November. Available at: https://www.linkedin.com/pulse/defining-ai-capability-holistic-framework-readiness-ai-driven-dixon-omrle [Accessed 4 September 2026].

Dixon, R.J., 2025. The AI Transformation Playbook: my first book, and why I wrote it. [online] robdixon.ai, 19 December. Available at: https://www.robdixon.ai/post/the-ai-transformation-playbook [Accessed 4 September 2026].

Dixon, R.J., 2025a. Where Does the Human Stop and the AI Begin? Purpose - Execution - Judgement (PEJ): A Framework for Human-AI Interaction in the Post-ChatGPT Era. [online] LinkedIn, 30 June. Available at: https://www.linkedin.com/pulse/where-does-human-stop-ai-begin-purpose-execution-judgement-dixon-bn0se [Accessed 4 September 2026].

Dixon, R.J., 2025d. The AI Transformation Playbook: A Step-by-Step Guide to Becoming an AI-Enabled Organisation. Manchester: Dixon AI Press. Available at: https://www.amazon.co.uk/dp/1919400508.

Dixon, R.J., 2026a. From Knowledge to Purpose: What Happens to Jobs as We Approach AGI. [online] LinkedIn, 8 February. Available at: https://www.linkedin.com/pulse/from-knowledge-purpose-what-happens-jobs-we-approach-agi-rob-dixon-vnqze/ [Accessed 4 September 2026].

Dixon, R.J., 2026b. Purpose, Execution, Judgement: A Framework for Human-AI Collaboration and Organisational AI Capability in the Post-ChatGPT Era. SSRN Working Paper. Available at: https://dx.doi.org/10.2139/ssrn.6282200 [Accessed 4 September 2026].

Hemmer, P., Schemmer, M., Vössing, M. and Kühl, N., 2025. Complementarity in human-AI collaboration: concepts, methodologies, and a research agenda. Information Systems Frontiers. Available at: https://doi.org/10.1007/s10796-024-10555-9.

Morrow, J., 2016. Shooting Ratios, From Hitchcock to 'Fury Road' to 'Primer' (and What They Mean to You). [online] No Film School, 4 March. Available at: https://nofilmschool.com/2016/03/shooting-ratios-mad-max-fury-road-primer-hitchcock [Accessed 4 September 2026].

Willison, S., 2024. Slop is the new name for unwanted AI-generated content. [online] Simon Willison's Weblog, 8 May. Available at: https://simonwillison.net/2024/May/8/slop/ [Accessed 4 September 2026].

Further reading. Added September 2026: later work that builds on this paper.

Dixon, R.J., 2026d. From Knowledge to Purpose: The Shifting Bottleneck of Value Creation and Five Domains of Human Value as We Approach AGI. SSRN Working Paper. Available at: https://dx.doi.org/10.2139/ssrn.6396778 [Accessed 4 September 2026].

Dixon, R.J., 2026e. AI Innovation Architecture: Nested Purpose, Closed-Loop Dynamics, and the Chief Agency Officer. SSRN Working Paper. Available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6546178 [Accessed 4 September 2026].

Dixon, R.J., 2026. From Knowledge to Purpose: what happens to human value when intelligence is cheap. [online] robdixon.ai, 23 March. Available at: https://www.robdixon.ai/post/from-knowledge-to-purpose-paper [Accessed 4 September 2026].

Dixon, R.J., 2026. AI Innovation Architecture: who sets purpose when your systems become agents? [online] robdixon.ai, 23 April. Available at: https://www.robdixon.ai/post/ai-innovation-architecture-chief-agency-officer [Accessed 4 September 2026].

Rob Dixon is Founder and CEO of Dixon AI Ltd and lectures at Manchester Metropolitan University. He writes and speaks on AI capability, strategy, and human-AI collaboration. Bibliography · Google Scholar · ORCID · LinkedIn · dixonai.com