Chris Dury
ai strategyproject managementportfolio managementstrategic executionai governanceoperating modeldecision intelligencesenior leadership

With AI, it's judgement that doesn't scale

AI turns execution into human-machine orchestration. Leaders must rethink projects, portfolios, governance and strategic visibility around judgement and trust.

10 May 2026 · Archive

A graphic representing a project manager's role in a human-machine orchestration system.

AI changes the constraint, not just the toolkit

The strategic question is no longer whether AI can help a project manager draft a status report, summarise a meeting, identify risks or assemble a first-cut plan. It can already do much of that, and those uses are often the first to scale because they absorb administrative and coordination work without immediately challenging the organisation's operating model.

The more consequential question is what happens when execution itself becomes partly machine-mediated. If AI can generate plans, update analysis, monitor signals, coordinate routine work and draft decision options, the human role shifts from direct task execution towards orchestrating, validating and governing human-machine work systems.

That is the change senior leaders need to understand. AI improves project management first, but the strategic implication is larger: it forces a rethink of the execution system. Projects do not disappear. They become one vehicle inside a broader architecture of portfolios, capabilities, platforms, value streams, AI-enabled workflows and governed decision environments.

The danger is to treat this as a project-management productivity upgrade. Faster reporting, cheaper analysis and better drafting are useful, but they do not automatically create strategic progress. They can just as easily multiply low-value work, make weak initiatives look more controlled, or overwhelm leaders with plausible but untested recommendations.

The scarce resource moves. Production becomes easier. Judgement, trust, coordination, accountability and strategic focus become more valuable.

For senior leaders, the practical question is therefore not: are our teams using AI? It is: can the organisation orchestrate AI-assisted execution safely, strategically and repeatedly?

The current landscape: augmentation is real, but it is not the endpoint

The current evidence points to two layers of change.

The first is the visible productivity layer. AI helps teams draft, summarise, classify, analyse, forecast and report. In project environments, that means faster status aggregation, risk identification, meeting documentation, schedule support, stakeholder communications and scenario analysis. Most early project-management adoption fits this pattern: familiar work is being augmented before the operating model is seriously challenged (AI project management; generative AI in project management).

The second layer is more structural. Once AI becomes embedded in workflows, decision processes and semi-autonomous agents, it changes how work is decomposed, delegated, monitored and integrated. The emerging pattern is not simply faster individual work; it is humans steering, revising, validating and supervising AI-supported workstreams while organisations redesign roles, governance and performance management around human-agent collaboration (future of work research; workforce orchestration; agentic AI in work).

This does not mean autonomous multi-agent orchestration is already the normal operating reality. ISG's evidence suggests that multi-agent deployments remain a minority of agentic solutions. The direction of travel is clearer than the current level of maturity. Most organisations are still closer to AI-assisted task work than to fully orchestrated human-agent operating systems.

Even so, the strategic signal is strong. AI is not only making existing project work faster. It is changing the balance between execution, supervision, judgement and governance. Once agentic AI is treated as both tool and coworker, leaders run into operating-model questions about autonomy, decision flows, roles, governance and KPIs rather than a simple process-improvement agenda (agentic enterprise research; agentic operating models).

The strongest conclusion is therefore nuanced. AI is currently improving established project-management practice. But as its use becomes more agentic, embedded and consequential, it pushes organisations towards a deeper redesign of project, portfolio and strategic execution.

The bottleneck moves from production to judgement

A useful way to understand the shift is to separate production capacity from decision capacity.

AI increases production capacity. It can create more plans, analyses, summaries, dashboards, options and recommendations. But decision capacity does not scale at the same rate. Senior judgement, stakeholder trust, legal accountability, organisational attention and implementation capacity remain limited.

This is the bottleneck shift: as AI produces more knowledge-work artefacts, human value moves towards intention-setting, taste, strategy, coordination, accountability and trust. That argument appears at the operating-model level, at the individual level, and in broader reviews of AI-mediated knowledge work (operating model; guidance economy; literature review).

Domain examples make the shift concrete. AI may compress M&A screening, valuation and due diligence, but negotiation, governance approval, accountability and integration still constrain deal quality. It can speed military mission analysis or research workflows, but the decisive work remains validation under uncertainty, expert challenge and responsible action (M&A analysis; military planning; AI-assisted research).

The same pattern applies to strategic execution. AI can generate more portfolio scenarios, but the executive team still has to choose. AI can produce more project intelligence, but someone must decide whether to act. AI can draft recommendations, but governance must establish whether those recommendations are reliable, traceable and aligned with strategy.

Faster production can therefore create the illusion of progress. A portfolio with continuously refreshed dashboards may still be strategically incoherent. A project with AI-generated reports may still be delivering the wrong thing. A strategy process with AI-generated options may still converge on generic choices if leaders do not inject context, trade-offs and distinctive judgement.

The better question is not how much more work AI can produce. It is whether the organisation can absorb, test, prioritise and act on AI-generated work responsibly.

Where the old project model breaks

The inherited execution model treats projects as the main unit of change. Strategy is translated into initiatives; initiatives become programmes and projects; projects are governed through scope, cost, schedule, risk, dependencies and benefits.

That model has been useful. Projects create accountability. They make investments visible. They time-box effort. They give leaders a way to allocate money and people to defined outcomes. The enterprise PMO becomes credible as a strategy execution office only because portfolios, programmes and projects can connect corporate strategy to measurable work when governed well (PMO as strategy execution office).

But the project-centric model also has familiar failure modes. Organisations confuse project volume with strategic progress. Initiatives continue long after the strategic case weakens. Portfolios become overloaded, executive attention is diluted and scarce capacity is fragmented. The useful corrective is not anti-project; it is a reminder that continuation should be an active strategic choice, not a default setting (focus on fewer projects).

AI intensifies these weaknesses. If planning and reporting become easier, organisations may launch even more initiatives. If analysis becomes cheaper, every function can generate its own AI-supported business case. If status reporting becomes automated, weak projects can look more controlled than they really are.

At the same time, AI value often depends on foundations that do not fit neatly inside isolated projects: shared data assets, reusable AI platforms, governance patterns, decision logs, human-AI workflows, operating-model redesign and enterprise learning. That is why isolated pilots are such a weak strategic unit. The deeper work is companywide capability building around business use cases and, in some contexts, rethinking the business model itself (AI-enabled organisations; business model rethink).

The old model breaks when a project is used as the container for work that is really about enduring capability. A project can build a data product, but the capability must be owned, improved, governed and reused. A project can deploy an AI agent, but the operating model must monitor it, update it, constrain it and hold someone accountable for its actions.

A better distinction: vehicle, capability, steering system and trust infrastructure

The previous problem is a category error: organisations treat projects as though they are the strategy, while much of AI-enabled value comes from capabilities, platforms, data foundations, decision systems and operating routines that outlive any single project.

A better distinction is this:

Projects are temporary vehicles. Capabilities are enduring assets. Portfolios are steering systems. Governance is the trust infrastructure.

A project remains useful when work needs a defined scope, budget, timeframe and accountable delivery team. The mistake is treating the project as the durable strategic unit. In AI-enabled organisations, the more durable unit is often a capability or value stream: a data asset, customer journey, operating process, product platform, decision system or human-AI workflow that continues to evolve after the project closes.

A portfolio is not merely a collection of active projects. It is a live set of strategic options competing for attention, funding, talent and risk capacity. If AI can refresh analysis continuously, portfolio management should become a steering discipline rather than a reporting ritual.

Governance is not a final approval gate. It is the set of decision rights, evidence requirements, controls, monitoring mechanisms and accountability structures that allow the organisation to rely on AI-assisted work without losing judgement.

This distinction changes the executive conversation. Instead of asking only whether a project is on track, leaders should ask:

  • What enduring capability is this project building or changing?
  • Does the portfolio still reflect our strategic priorities?
  • What work should be stopped, paused or reallocated?
  • Which AI-generated recommendations are being used, and under what controls?
  • Who is accountable for decisions influenced or executed by AI?
  • Can we reconstruct the evidence, data and human judgement behind important decisions?

The shift is from project control to execution architecture.

Project management becomes orchestration and accountable sense-making

In this new model, project management becomes less about task control and more about orchestration, verification and accountable sense-making.

That does not make project management less important. It makes it more judgement-intensive. As AI takes on routine drafting, reporting, scheduling support, risk scanning and analysis, the project leader's comparative value shifts to designing the conditions under which human and machine work can be safely combined.

That includes:

  • designing workflows in which humans and AI each do appropriate work;
  • setting guardrails for AI-generated plans, analyses and communications;
  • validating outputs before they influence stakeholders or decisions;
  • managing exceptions, ambiguity and escalation;
  • preserving context across teams, tools and agents;
  • ensuring accountability is explicit when AI contributes to project work.

The underlying shift is a change in what expertise is for. When AI democratises access to answers, human value moves towards context, judgement and accountability. Project planning, reporting and risk management may become more dynamic, but roles, governance and decision-making still need to be rethought rather than discarded (expertise in the age of AI; AI project management; digital transformation synthesis).

The project manager becomes a sense-maker, verifier and accountable integrator. They must understand the work well enough to challenge AI outputs, and the system well enough to decide where automation helps and where it creates risk.

This creates a talent problem. If AI absorbs the entry-level work through which people traditionally learned to analyse, document, test assumptions and develop professional judgement, organisations will need new development pathways. Agent-led orchestration implies redesigned tasks, roles, teams and workforce planning, but the evidence does not yet resolve how the next generation of senior project, product and strategy leaders will build judgement if the apprenticeship layer of knowledge work is hollowed out (workforce strategy).

For leaders, that is not a marginal HR issue. It is a strategic execution risk. Judgement cannot be delegated to AI if the organisation stops developing the humans responsible for judgement.

Portfolio management becomes continuous steering

Portfolio management is where the change becomes most visible at executive level.

Traditional portfolio management often operates through periodic cycles: annual planning, quarterly review, monthly reporting, stage gates and steering committees. AI challenges that cadence. If plans, forecasts, risks and scenarios can be updated more dynamically, waiting for the next reporting pack becomes harder to justify.

The direction of travel is continuous, AI-supported steering: plans, analysis and resource allocation updated close to the pace at which evidence changes. Dynamic forecasting and scenario analysis matter because traditional planning outputs can become obsolete too quickly for fixed annual cycles to remain the main steering mechanism (AI-supported portfolio steering; dynamic financial steering).

The practical change is not simply faster reporting. It is a different job for the portfolio forum. Instead of asking teams to explain stale status data, executives should use portfolio governance to make frequent, evidence-based decisions about strategic fit, sequencing, funding, capacity, risk appetite, dependencies, benefits realisation and exit, pause or pivot decisions.

For the PMO or portfolio office, this is a role shift as much as a tooling shift. The centre of gravity moves away from compliance-heavy consolidation and towards strategic enablement, shorter prioritisation cycles, integrated data, real-time collaboration, dynamic resource allocation and value-based decision criteria. Platforms matter because strategy capture, roadmapping, financials, capacity planning and AI-driven analytics have to be connected before portfolio decisions can become genuinely data-driven (AI-powered portfolio management; strategic portfolio capabilities). A mission-control model for pharmaceutical R&D shows the same logic in a high-complexity setting: the task is to turn fragmented dashboards into continuous, AI-enabled portfolio oversight (R&D mission control).

Dynamic management should not become constant churn. The point is not to change priorities every time a new signal appears. It is to maintain a live view of strategic options and move funding, people and attention when the evidence justifies it.

That requires clearer decision rights, not looser ones. AI can draft scenarios and recommendations. Leaders remain accountable for interpreting signals, testing assumptions and deciding.

Senior visibility shifts from control reporting to decision intelligence

Senior leaders will need a different visibility layer.

Traditional project controls answer important questions: Are we on time? Are we on budget? What are the risks? Which dependencies are blocking progress? Those questions remain necessary, but they are no longer sufficient.

The next visibility layer needs KPIs that are descriptive, predictive and prescriptive, and that reveal relationships among metrics across silos. That reframes executive visibility as decision intelligence, not just reporting (AI-enhanced KPIs).

Leaders need to see:

  • leading indicators, not only lagging status;
  • cross-functional drivers, not only project-level metrics;
  • strategic trade-offs, not only delivery health;
  • AI recommendations and their rationale, not only outputs;
  • data lineage and model trustworthiness, not only dashboard results;
  • workflow redesign, including where agents act and where humans intervene;
  • benefits realisation, not only implementation progress;
  • sustainability, ethics and legitimacy impacts where material.

The dashboard therefore has to become more than a prettier status pack. Useful executive visibility is real-time, cross-functional, predictive and explainable; it shows value creation, workflow redesign, governance, workforce change, orchestration and sustainability impacts as well as traditional delivery health. At board level, the same logic becomes a governance question: AI adoption is outpacing oversight, so leaders need visibility into data readiness, adoption pace and emerging risk exposure (decision intelligence systems; AI predictions; board governance).

Leaders should also govern the measurement system itself. Even top teams can lack a shared understanding of strategy, which makes a small number of clear priorities, frequent communication and translation into focused goals more important. AI will not fix unclear strategy; it will amplify it (strategic alignment; strategic agility).

A green dashboard can still be a strategic failure if it measures the wrong work.

Governance becomes continuous, risk-tiered and embedded

When AI increases the speed, volume and apparent confidence of outputs, governance must change.

Traditional governance often relies on human review, document sign-off, IT approval or compliance checkpoints. Those mechanisms are too slow and too coarse for AI-mediated knowledge work. They also focus too much on the output and not enough on the lifecycle that produced it.

The common governance pattern is lifecycle-based and risk-tiered. Leaders need to manage AI across design, development, use, evaluation and retirement, with inventories, risk ratings, documentation, testing, human oversight, ownership and monitoring that become stronger as systems become more complex or consequential (AI Risk Management Framework; model risk management).

For senior leaders, the practical governance model should have six layers:

  1. Enterprise accountability: board and executive oversight, named owners and clear accountability for AI systems and AI-assisted decisions.
  2. Risk-tiering: different controls for low-risk productivity use, high-impact decisions and autonomous or agentic workflows.
  3. Lifecycle management: governance from design and data sourcing through deployment, monitoring, updating and retirement.
  4. Embedded controls: approved platforms, data access rules, audit logs, automated policy checks and secure environments.
  5. Continuous assurance: monitoring, testing, drift detection, incident pathways, revalidation and independent review where needed.
  6. Human judgement rules: explicit thresholds for when humans must review, challenge, approve or override AI outputs.

The controls need to live inside the operating system of work: process controls, testing, auditing, data governance, certification, cross-functional accountability and post-deployment monitoring. After-the-fact review will be too late for AI-assisted execution that is already shaping decisions (AI risk mitigations; enterprise AI governance).

The broader literature adds the same caution from different angles: responsible AI needs lifecycle design, auditability and accountability; generative AI raises familiar risks around bias, transparency, misuse and human oversight; trustworthy AI in high-stakes domains depends on stakeholder-specific requirements, domain knowledge and lifecycle controls (responsible AI framework; trustworthy AI decision-making).

The leadership lesson is simple: governance that waits until the output is complete will not keep pace. Governance has to be built into the production of knowledge itself.

Trust depends on decision rights and provenance

Accountability is the linchpin of AI-enabled execution.

In traditional execution, accountability usually attaches to people and roles: sponsor, project manager, product owner, accountable executive, steering committee. That remains necessary, but it becomes insufficient when AI systems recommend, prioritise, draft, evaluate or initiate work.

AI copilots create value only when organisations redesign workflows, decision rights and governance around human judgement rather than treating AI as a simple add-on. That extends into a broader choice-architecture problem: leaders must shape the decision environment, not just govern the model (AI copilots and accountability; intelligent choice architectures).

That means leaders need decision charters for high-impact AI-assisted execution. These should specify:

  • when AI may advise;
  • when AI may act within guardrails;
  • when human review is mandatory;
  • who may override AI recommendations;
  • what evidence must be documented;
  • how exceptions are escalated;
  • how contested decisions are reviewed;
  • what audit trail must be preserved.

Provenance also needs to broaden. It is not only data lineage. It is the chain of custody for an AI-assisted decision: data sources, model assumptions, prompts or inputs, AI recommendations, human changes, rejected alternatives, approvals, overrides and resulting actions.

Transparency, accountability, traceability and risk management become harder as systems gain agency. Autonomous systems may act, create identities or modify infrastructure without clear human approval, so acceleration has to be paired with governance, human oversight and risk controls if trust is to survive (AI principles; auditing agentic AI; AI governance memorandum).

The conclusion for executives is uncomfortable but necessary: accountability cannot be delegated to an agent. It must be designed into the decision environment before the agent acts.

What organisations should build now

Many organisations are asking where to invest first. The answer is not another tool rollout. It is a connected capability system.

The pattern in scaled AI work is consistent: pilots do not become enterprise value unless organisations build strategy, systems, people, data, governance and leadership ownership around redesigned workflows (AI maturity capabilities; state of AI; AI survey).

For project, portfolio and strategic execution, seven capabilities matter most.

1. Strategic AI execution discipline. Leaders need the ability to connect AI investments to strategic priorities, portfolio choices and measurable value. LLMs can support strategy generation and evaluation in specific contexts, but human strategists and complementary assets remain critical (strategy and LLMs).

2. Human-AI workflow design. Organisations need to decompose work, allocate tasks between people and AI, define escalation paths and redesign roles around outcomes. This is the practical meaning of human-machine orchestration: shared decision rights, digital trust and continuous adaptability (human capital trends).

3. Data, platform and integration foundations. Portfolio and strategy decisions require connected data across finance, projects, operations, customers, people and external signals. Without this foundation, AI accelerates poor information.

4. Workforce AI fluency. People need role-specific AI literacy and enough in-house capability to redesign work safely rather than merely automate old processes (AI-ready workforce).

5. Governance and digital trust. Validation, audit trails, risk review, data protection, compliance and explicit decision rights must be treated as execution capabilities, not compliance overhead. Responsible AI blueprints, GRC research and AI-driven management reviews all point to accountability, transparency, assurance and risk management as central to trustworthy adoption (Utah RAI; Nature; ScienceDirect).

6. Organisational agility and resource orchestration. AI-mediated execution requires faster sensing, experimentation, scaling and reallocation. AI capabilities appear to support value definition, creation and capture most effectively when organisations also build agility, learning and governance capabilities rather than merely acquiring tools (project-based organisations; manufacturing SMEs).

7. Benefits realisation and measurement governance. Leaders need proof that AI-enabled execution creates value, and they need to know when it is creating risk, dependency, waste or capability erosion. Productivity gains may correlate with organisational performance, but that signal still has to be translated into governed strategic value (AI and productivity).

The most mature organisations will not be those with the most AI experiments. They will be those that can repeatedly turn AI-enabled insight into governed, strategically aligned action.

Implications for senior leaders

Senior leaders should make several practical shifts.

Stop treating AI as a productivity sidecar. Individual productivity gains are useful, but they do not automatically create enterprise value. AI must be connected to workflows, portfolios, governance and strategic priorities.

Stop using projects as the default container for every strategic idea. Projects remain useful, but leaders should ask what enduring capability, platform, process or value stream the project is building. If there is no durable strategic asset, the project deserves harder scrutiny.

Stop rewarding reporting volume. AI will make it easier to produce reports. That makes it more important to reward decision quality, issue resolution, learning, benefits realisation and strategic focus.

Start managing the portfolio as a live set of options. Funding and capacity should move when evidence changes, but within clear decision rights and strategic guardrails.

Start measuring AI trustworthiness, not just AI usage. Usage without assurance can scale risk. Leaders need metrics for validation, model performance, incidents, override patterns, data quality, auditability and decision provenance.

Start redesigning talent pathways. If AI absorbs routine execution, organisations need deliberate ways to build judgement. That may mean more simulation, review apprenticeships, rotational exposure, expert critique, structured exception handling and explicit evaluation of judgement quality.

Start clarifying human accountability for machine-mediated action. Every high-impact AI-assisted workflow should have named human owners, escalation rules, evidence standards and audit trails.

Start broadening strategic criteria. AI-enabled execution is not only about speed and productivity. Depending on context, leaders may also need to consider sustainability, stakeholder accountability, ethics and legitimacy. AI can reshape internal execution, stakeholder expectations, ecosystem boundaries and assurance obligations at the same time (ESG and AI capability; digital transformation; stakeholder accountability).

The central leadership shift is from asking whether teams are using AI to asking whether the organisation can orchestrate AI safely, strategically and repeatedly.

What good looks like

A mature response would have several visible characteristics.

First, strategy would be simpler and better understood. AI cannot compensate for strategic ambiguity. It will amplify ambiguity by generating more plausible plans, metrics and options around priorities that leaders have not genuinely aligned on.

Second, the portfolio would be actively pruned. Leaders would stop low-value work faster, not merely launch AI-enabled initiatives faster.

Third, the PMO or portfolio office would become an orchestration and decision-intelligence function. It would maintain integrated data, scenario views, capacity signals, benefits tracking and governance discipline. It would not be primarily a reporting factory.

Fourth, AI agents and copilots would be governed through risk tiers, inventories, monitoring and clear accountability. High-impact workflows would have decision charters and provenance requirements.

Fifth, project teams would be trained to work with AI as a collaborator whose outputs must be checked, contextualised and integrated. Teams would learn not only prompting, but verification, challenge, escalation and judgement.

Sixth, executive dashboards would show both performance and trust. Leaders would see not only what AI recommends, but what data it used, what limitations apply, who reviewed it and what residual risks remain.

Seventh, innovation work would become more iterative and ecosystem-based. AI shifts innovation away from a purely linear process and towards dynamic human-AI ecosystems in which managers orchestrate recombination, targeting and continuous improvement (generative AI and innovation).

Finally, the organisation would distinguish between speed and progress. Faster analysis is valuable only when it improves strategic choices, accelerates learning, reduces waste or strengthens execution.

The organisations most likely to benefit are not those that bolt AI-generated reporting onto static annual governance. They are the ones that redesign decision processes, behaviours, roles and controls around dynamic information.

What to watch next

Several signals will show whether the shift is becoming real.

Watch whether portfolio forums change their agendas. If executives still spend most of their time reviewing status packs, the operating model has not changed. If they spend more time making trade-offs, reallocating capacity, stopping weak work and testing scenarios, portfolio management is becoming continuous steering.

Watch whether AI governance moves into workflow infrastructure. Policies alone will not be enough. The real test is whether approved data sources, access controls, audit logs, risk tiers and monitoring are embedded in the tools and processes people use every day.

Watch whether PMOs evolve. A PMO that remains focused on compliance and consolidation may be displaced. A PMO that becomes a strategic enablement, decision-intelligence and governance function becomes more important.

Watch the talent pipeline. If junior roles are reduced without replacing the learning they provided, organisations may enjoy short-term productivity while weakening long-term judgement.

Watch whether leaders ask better questions. The most important executive questions will sound less like status control and more like strategic orchestration:

  • What decision is this AI output meant to improve?
  • What evidence would cause us to stop or pivot?
  • What human judgement has been applied?
  • What risks are hidden by the apparent confidence of the output?
  • What capability are we building that persists beyond the project?
  • Can we reconstruct the decision if challenged?

These signals matter because the future is not evenly distributed. Some organisations will use AI to automate old bureaucracy. Others will use it to build a more adaptive execution system.

Where I land

AI does not make project management obsolete. It makes traditional project-centric execution insufficient.

Projects remain necessary because organisations still need bounded delivery, investment control, accountability and coordination. But projects should no longer be treated as the primary expression of strategy. In an AI-enabled organisation, the more durable unit is the strategically governed capability or value stream, supported by reusable data and AI platforms, human-AI workflows, clear ownership and continuous measurement.

Project management becomes the discipline of orchestrating human-machine work. Portfolio management becomes the discipline of continuously steering strategic options. Strategic execution becomes the discipline of building and governing capabilities that can learn, adapt and create value over time.

The durable insight is that AI relocates scarcity. It makes production less scarce, but judgement, trust, coordination, accountability and strategic focus more valuable.

Senior leaders should therefore avoid two mistakes. The first is underestimating AI by treating it as a reporting assistant. The second is overestimating AI by assuming it can own judgement, accountability or strategic intent.

The right posture is disciplined ambition: redesign execution around AI, but keep humans accountable for the choices that matter.

References