I’ve given this talk before
In 2016 I stood on stage at Adaptive Path’s MX conference and argued that design organizations are complex adaptive systems, not complicated machines. My warning was that if you manage a complex organization as if it were just a complicated one, you’ll make serious, expensive mistakes. I borrowed Irene Ng’s line to make the point: “We can only intervene in complex systems.”
Three years later, at Interaction 19, I opened with Klaus Schwab’s Fourth Industrial Revolution: mobile internet, automation and artificial intelligence. AI was the third item on a slide. In 2026 it is the whole slide.
The pressure is new. The shape of the problem is not. Google’s 2025 DORA report, based on nearly 5,000 technology professionals, concludes that AI’s primary role in software development “is that of an amplifier.” Strong teams get stronger; struggling teams get their dysfunctions magnified. That is the complex-systems argument I made a decade ago, now with survey data behind it.
Donella Meadows described leverage points as places where “a small shift in one thing can produce big changes in everything” (Meadows). She also observed that people usually intervene at the weakest points: parameters, rates and numbers. Buying AI tools is a parameter change. The organizations seeing real returns are changing structure, and design and leadership sit at the center of that work.
Six shifts, revisited
From 2019 to 2023, I built talks around Leyla Acaroglu’s “Tools of a System Thinker,” six moves from a mechanical view of work to a systemic one (Acaroglu). I paired each with behaviors I asked of design leaders. AI amplifies four of the six and complicates the other two.
| Shift | What I asked of leaders in 2019 | What AI does to it in 2026 |
|---|---|---|
| Disconnection → interconnectedness | Set a clear north star; involve diverse teams sooner | Amplifies it. Agents work across tools and teams at once, so a weak north star now produces divergent output at machine speed. |
| Linear → circular | Map experiences beyond the perceived start and end of use | Amplifies it. AI products change through use, so the loop is the product. |
| Silos → emergence | Watch behavior over time through multiple feedback loops; welcome unexpected outcomes | Complicates it. Emergence now comes from the model too, and some outcomes were never designed by anyone. |
| Parts → wholes | Talk to secondary and tertiary roles; map the wider ecosystem | Amplifies it. Designers now define agent goals, escalation paths and guardrails across that ecosystem. |
| Analysis → synthesis | Let the group evaluate individual ideas; practice double-loop learning | Complicates it. AI synthesizes research in seconds; the risk is outsourcing judgment about which insights matter. |
| Isolation → relationships | Build psychological safety and cognitive humility | Amplifies and extends it. The team now includes agents people supervise, not just colleagues. |
The 2019 framing still holds. What changed is that each shift now runs at a speed and scale where getting it wrong compounds quickly.
The constraint moves
When one part of a system speeds up, the bottleneck moves somewhere else. Andrew Ng has argued that because AI made coding dramatically faster, deciding what to build is now the limiting step (Business Insider). A year later he went further: when software gets 10 to 100 times faster to write, “pretty much everything else becomes a bottleneck” (interview transcript).
The constraint also moves downstream. DORA’s lead, Nathen Harvey, wrote that AI-driven acceleration tends to expose weak spots later in the delivery pipeline (Google Cloud). DX’s analysis of the same data points to code review as the step most likely to choke when code volume rises (DX).
The shape of the process changes too. McKinsey notes that agents work asynchronously and around the clock, so development no longer has to pass through a linear chain of human handoffs (McKinsey). A process designed around handoffs will fight a system that no longer has them.
This is the difference I drew in 2016 between complicated and complex systems. In a complicated system you can predict where the bottleneck goes next. In a complex one it moves somewhere you didn’t plan for.
That talk also argued that longer, looser feedback loops let disruptions go unnoticed. I urged design managers to close them by inviting executives into design studios, sharing research with everyone, and keeping Support, Sales and Marketing in the room. AI makes that advice urgent. When building takes hours, a feedback loop measured in weeks means shipping the wrong thing several times before anyone notices.
The product becomes a system too
In 2016 I quoted Sharon VanderKaay’s description of complex adaptive systems: diverse, interconnected agents with the capacity to grow and change. I meant people, teams and partners. In 2026 the word “agents” is literal, and it describes the product as much as the organization.
AI changes what teams build, not only how. Large language models are probabilistic, so designers can no longer map every path in advance. Thoughtworks argues the job becomes keeping an experience coherent whatever the model returns (Thoughtworks).
The work shifts from screens to behavior. A designer on an agentic product spends less time on layouts and more on handoff criteria, confidence thresholds and fallback messaging. One uxdesign.cc essay frames this as moving from designing interfaces to designing relationships and environments, because AI systems change through use (uxdesign.cc).
That requires new measures of health. Thoughtworks suggests tracking user confidence, how often people intervene, and how often they override the system. Screen-level metrics like task completion miss whether the system as a whole is earning trust.
The role of design: architect the system
Design’s value moves from producing artifacts to shaping the system that produces them. Thoughtworks lists systems thinking first among the capabilities designers now need, summed up as “Design ecosystems instead of individual screens.” The same piece names context engineering: defining the information and decision rules that shape how an AI behaves.
The design system becomes infrastructure. Luke Wroblewski describes a design system built for agents as “a steering layer for a site’s design and front-end code” (LukeW). When agents write most of the code, the system is what keeps output on-brand and consistent.
This is a straight line from my 2019 advice to scaling design leaders: use the design system as a bridge between teams, and position governance as a way to avoid problems later, not a power grab. The bridge is the same. The traffic is new: it now carries agents as well as designers and engineers.
That raises the bar for system quality. Frances Liddell puts it plainly: “A design system that isn’t reliable for people, won’t be reliable for AI” (Liddell). Mismatched tokens and stale documentation become errors that agents reproduce at scale.
There is a risk to the craft itself. Kike Peña warns that when fast output becomes the only quality standard, the holistic thinking design depends on starts to look unnecessary (uxdesign.cc). Protecting time for problem framing is a design leadership job, not a personal habit.
The role of leadership: change the structure
Leaders get returns from AI by redesigning the operating model, not by adding tools. McKinsey’s August 2026 research is blunt: “Technology is not the differentiator.” What separates leading organizations is how they adapt end-to-end workflows, roles, controls and change management to an agentic lifecycle.
The gap between belief and results is wide. A 2025 Harvard Business Review Analytic Services study found 83% of product decision-makers think AI can significantly improve development, but only 34% are getting the results they want (as reported by Monterail).
Team composition has to follow the new constraint. Ng expects product and design work to grow as a share of the workforce as coding gets cheaper (HackerNoon). Staffing ratios built for slow engineering will starve the steps that now limit throughput.
Meadows’ hierarchy gives leaders a checklist. Low leverage: licenses, velocity targets and usage dashboards. High leverage: what the organization measures, who sees which information, the rules for what can ship, and the goals teams are rewarded against.
My 2016 talk leaned on the adaptive cycle: rapid growth, conservation, release and reorganization. Late in the conservation phase, organizations chase efficiency by stripping out redundancy, tightening command and control, and suppressing experimentation. Many AI programs push in exactly that direction: fewer people, tighter process and more measured output.
The resilience moves I recommended then still apply, each with an AI-era form:
Success measures need the same update. In 2019 I contrasted fixed-goal questions, such as whether the boss liked it or whether we handed off on time, with systemic ones: did it create business value, and what did we learn that we must act on now? When output is cheap, “did we ship it” measures almost nothing.
Open tensions
- Speed versus coherence. When anyone can ship, who owns the system that keeps the product consistent?
- Local versus system gains. Individual productivity numbers can rise while end-to-end throughput stays flat.
- Judgment under uncertainty. Who defines “good” when the product’s behavior is probabilistic?
- Who benefits. AI may widen the gap between strong and weak teams rather than close it.
- Efficiency versus resilience. AI invites late-conservation thinking: cut redundancy, tighten control, measure output. That is the phase that precedes collapse.
What changed, and what didn’t
What hasn’t changed: organizations are complex adaptive systems, leaders can intervene but not determine, and feedback loops decide whether surprises are caught in time.
What has changed: the agents in the system now include software, the loops run at machine speed, and the design system has become the steering layer.
I closed the 2016 talk by urging leaders to make changes from a position of strength, not desperation. That advice matters more now. The pressure to adopt AI pushes many organizations toward desperation, and desperate systems intervene at the weakest leverage points.
Sources
- DeBellis, D. et al. State of AI-assisted Software Development. DORA, Google, 2025.
- Harvey, N. Announcing the 2025 DORA Report. Google Cloud, September 23, 2025.
- Meadows, D. Leverage Points: Places to Intervene in a System. Donella Meadows Project, 1999.
- Lee, C. M. Andrew Ng says the real bottleneck in AI startups isn’t coding. Business Insider, August 2025.
- Ng, A. Interview transcript. CEOInterviews.ai, 2026.
- Andrew Ng: Product team ratios evolving. HackerNoon.
- DORA’s 2025 research on the impact of AI. Engineering Enablement, DX.
- Beyond the copilot: Scaling the agentic product development life cycle. McKinsey, August 21, 2026.
- Product design in the age of AI: Designing non-deterministic systems. Thoughtworks.
- Beyond the Model: a systems approach to AI product design. UX Collective.
- Peña, K. Did AI kill the system’s thinking skills in UX? UX Collective.
- Wroblewski, L. Design Systems for AI Agents. LukeW, October 1, 2026.
- Liddell, F. A design system that isn’t reliable for people won’t be reliable for AI. October 1, 2025.
- AI in Product Development Life Cycle. Monterail, citing Harvard Business Review Analytic Services, September 2025.
- Avore, C. Resilient Design Management: Embracing Complexity in Teams, Products, and Organizations. Adaptive Path MX conference, 2016. Includes quotations from Irene Ng and Sharon VanderKaay.
- Avore, C. The Future of Design Leadership: Transforming Design Management to Thrive in the Fourth Industrial Revolution. Interaction 19, IxDA, February 2019.
- Acaroglu, L. Tools for Systems Thinkers: The 6 Fundamental Concepts of Systems Thinking. Disruptive Design, September 7, 2017.