ICYMI
• The Plausibility Crisis – As AI-generated output becomes more abundant, convincing and unreliable, senior executives are becoming the last line of defence.
TL;DR
• Conway’s Law says that organisations build IT systems that copy their communication structures.
• AI is reversing the direction of travel: AI systems are reshaping organisations.
• The shift is happening through changed workflows, invisible human-agent teams, digital labour, delegated authority, culture, strategy and talent pipelines.
• A reversed Conway effect brings benefits such as increased efficiency and agility but also risks divergence between the organisation you have and the organisation you want.
• Organisations should treat AI as operating-model redesign, not as a technology rollout.
In 1968, the computer scientist Melvin Conway observed that any organisation that designs a system “will produce a design whose structure is a copy of the organization’s communication structure.” This became known as Conway’s Law: software mirrors the organisation that made it. If teams don’t communicate well, then their systems won’t either. Decentralised business decision-making, without strong technology governance, leads to fragmented IT architecture. The org chart leaks into the code.
AI is now reversing the arrow. As agents gain access to enterprise data, workflow tools, codebases and calendars, they do not simply mirror the organisation. Rather, they start to redesign it. They route work, summon other agents, trigger actions, and direct human effort, each of which changes how the organisation functions. We can call this Conway’s Reversal.
This is not true everywhere yet. Many deployments remain narrow: chatbots, copilots and summarisation assistants. But the direction is clear. Microsoft has described the rise of the “Frontier Firm”, built around hybrid teams of humans and agents. Its 2025 Work Trend Index found that 81 per cent of leaders expected agents to be moderately or extensively integrated into their AI strategy within 12 to 18 months.
The board-level question is not whether AI will change the organisation (it will) but whether the change is conscious. Firms can wake up to find that their workflows, teams, labour model, authority structures, culture, strategy and leadership pipeline have all shifted through accumulated AI adoption. Or they can purposely enlist AI to help design, build and operate an improved organisation.
Seven Quiet Redesigns
The Workflows No One Owns
AI is being inserted into workflows faster than those workflows are being consciously redesigned. A process that once had clear human handoffs becomes a chain of people, agents, integrations and exceptions. The process being followed quickly diverges from approved standard operating procedures.
The Teams Off the Org Chart
AI can dynamically assemble working groups that don’t appear on the org chart. A customer issue may involve a service agent, a pricing agent, a legal-review agent and a customer-success manager. That is a team, but it has no reporting line. It may assemble for a task and disappear once the work is complete.
The org chart is not dead, but it diverges from how work gets done. The practical unit of organisation may become the workflow, the customer journey or the agent permission boundary.
The Hires You Didn’t Make
The AI workforce does not arrive through hiring. It arrives through procurement, SaaS renewals and innovation budgets. The firm may not have approved new headcount, but it has added productive capacity. At IKEA’s largest retailer, Ingka Group, an AI assistant called Billie resolved 47 per cent of the customer enquiries it received between 2021 and 2023, while 8,500 call-centre workers were retrained as remote interior design advisers. Such agents are not employees in a legal sense, but they perform work, consume budget, and require oversight.
The Authority You Quietly Delegated
To be effective, agents need to be granted authority to (for example) recommend, approve, refund, reject, prioritise or contact customers. Access rights can quietly become decision rights if humans don’t challenge AI decisions. Any system that prioritises leads, complaints or candidates is making organisational judgements.
The Culture Your AI Normalises
Culture is not what leaders say, it is the repeated behaviours of the organisation. If AI handles repeated interactions, it starts to normalise (for example) tone, pace, empathy, caution, escalation and standards of judgement. A company may say it values customer intimacy while its agents optimise for speed. It may say it values judgement while training people to accept machine outputs.
The Strategy You Didn’t Approve
AI systems optimise local goals: increase conversion, lower cost, maximise throughput. Over time, the sum of those local optimisations can effectively become the company’s strategy. For example, a board may approve premium service while agents are tuned for efficiency. It may approve innovation while the controls around its AI reward caution.
The Leaders You Aren’t Developing
AI may automate the work through which people learn judgement. Junior employees have historically developed by drafting, analysing, coordinating, making mistakes and watching senior people. If AI absorbs that apprenticeship layer, the leadership pipeline changes.
The answer is not to stop this
Conway’s Reversal is not inherently bad. Handled well, it can drive productivity gains and increase organisational agility. Historically, software froze yesterday’s silos into today’s systems. The reversal lets a company shape the organisation that best suits the work and reshape it as circumstances change. Coordination that used to run through meetings now passes between agents directly, accelerating decision-making, and a small team directing a fleet of agents can produce what once took a large team. As routine work is absorbed, people are freed to move up towards higher-value tasks. The challenge for leaders is to ensure that AI evolves the organisation purposefully rather than by stealth. The right management response is therefore not to apply the brakes, but instead to make the new operating model visible, governable and strategically useful.
Making the Redesign Deliberate
1. Make the AI-shaped organisation visible. Executives need a live map of where AI is already changing work, including embedded SaaS features and agentic workflows. The aim is to see the real organisation, not the one implied by the org chart.
2. Redesign workflows, not isolated tasks. Value comes from redesigning work end to end, not sprinkling AI over old processes. Management should pick the priority workflows (for example, customer onboarding) and decide deliberately what humans do, what agents do, and what gets escalated.
3. Define authority, accountability and controls. Every material AI use case should be classified by its level of authority: inform, draft, recommend, decide, execute or escalate. Controls should then match the risk. A transcription agent does not need the same scrutiny as one that issues refunds.
4. Treat digital labour as workforce strategy. Agents and copilots should be part of workforce planning. Management should determine where AI substitutes for human work, where it augments employees, where it creates new supervision tasks and where it threatens apprenticeship. HR, finance, procurement, technology and risk need one shared view.
5. Align AI with strategy and culture. Local AI optimisation can quietly redirect the firm. Executives should check whether AI systems reinforce the approved strategy and the desired culture. If the strategy is premium service, agents cannot be tuned for cost.
Monday Morning Actions for Executives
• Ask for a one-page map of the most material AI-enabled workflows.
• Select pilot workflows and owners to redesign them around humans and agents.
• Classify material AI uses as inform, draft, recommend, decide, execute or escalate.
• Inventory agents and non-human identities, including credentials, permissions and system access.
• Ask HR to identify which apprenticeship tasks AI is removing and how judgement will now be developed.
Board Questions for the Back Pocket
• Which of our workflows is AI reshaping, and who owns them end to end?
• Where have we added digital labour, and what human work is it replacing, augmenting or supervising?
• What can our AI systems decide, recommend or execute, and where does human accountability sit?
• Are our AI systems reinforcing the strategy and culture we approved, or optimising towards something else?
Conway showed that organisations build software in their own image. AI has reversed the picture. The software is now rebuilding the organisation, one deployment at a time, and mostly by default. The org chart is being redrawn either way. The question is whose hand is on the pen?
Sources & Notes
Source caveat. Several figures below come from vendor reports or company-reported claims. They should be read as directional evidence of how the market is moving, not as benchmarks.
• Conway’s Law. Melvin Conway, “How Do Committees Invent?”, Datamation, April 1968. The original formulation: “Any organization that designs a system (defined broadly here) will produce a design whose structure is a copy of the organization’s communication structure.” The paper is reproduced on Conway’s own site. source
• Frontier Firm. Microsoft, 2025 Work Trend Index, “2025: The Year the Frontier Firm Is Born”, April 2025. The survey found 81 per cent of leaders said they expected agents to be moderately or extensively integrated into their company’s AI strategy within the next 12 to 18 months. Microsoft sells Copilot and agent tooling, so the framing serves its commercial narrative. source
• Ingka/Billie. Ingka Group newsroom and Reuters, June 2023: Ingka, the largest IKEA retailer, said its AI assistant Billie, launched in 2021, resolved approximately 47 per cent of the customer enquiries it received between 2021 and 2023, some 3.2 million interactions, while 8,500 call-centre co-workers were reskilled as remote interior design advisers. Company-reported figures, so directional rather than independently audited. source