
Sign up to save your podcasts
Or


The ability to absorb AI rests largely on how an organisation leads change.
In this ADAPT Insider episode, David Walker, former Group Chief Technology Officer at Westpac and DBS and ADAPT Advisor, draws on his experience transforming DBS to explain how an ambitious purpose gives people permission to innovate throughout the organisation.
He also explores how enduring characteristics can guide technology strategy without locking the business into platforms that may quickly date.
Key takeaways:
A shared purpose gives transformation direction
Organisational nimbleness is the ability to change and reinvent when external conditions shift.
Technology can support that movement, but leadership determines whether people have the direction and authority to act.
During the transformation of DBS, fear of disruption initially supplied urgency.
The organisation later adopted a more ambitious goal: becoming the world’s best bank.
That purpose gave employees permission to think differently.
Innovation extended beyond a specialist team as people throughout the business learned the skills to develop new ideas and act closer to customers.
David says the CEO and executive team must visibly own the ambition. Consistent communication from senior leaders connects everyday decisions to the organisation’s purpose.
Organisational barriers restrict AI absorption
David’s research identified more than 1,300 barriers to AI across 105 countries.
Around 70% related to organisational factors such as leadership alignment, processes, training and finance systems.
About 9% related directly to AI technology.
Many organisations direct their investment towards that smaller technical category.
New tools may increase access to AI, yet they cannot resolve unclear ownership, inflexible processes or workforce capability gaps.
AI adoption measures whether people have access to the technology.
AI absorption reflects whether the organisation can redesign work around it and convert that access into sustained value.
Smaller and mid-sized organisations may hold an advantage because fewer layers can allow them to change direction faster.
Their opportunity lies in building the leadership alignment, learning culture and operating flexibility required to challenge established organisations.
Enduring characteristics protect strategic flexibility
Fast-moving technologies can make platform-led strategies vulnerable to technical debt and lock-in.
A technology selected today may constrain the organisation as the market evolves.
David recommends defining the destination through enduring characteristics.
Systems might be automated, built for change, organic, digital to the core and evergreen.
These characteristics give boards, employees and vendors a shared view of the intended environment without prescribing a fixed technology stack.
Deliberate abstraction layers also allow organisations to replace AI models as new options emerge.
Strategic flexibility comes from designing systems that accommodate change before individual technologies harden into long-term constraints.
Every production agent should report its outcomes, costs, quality drift, failures and owner into one operational view.
In Episode 3 of ADAPT Insider’s AI Economics Series, The Control Tower and Monday Morning Questions, Vijayan Seenisamy, Author of The Pilot Trap | Enterprise AI Delivery, continues his conversation with ADAPT's Head of Programs and Value Engagement Byron Connolly to outline a practical model for governing AI agents in production.
The final episode covers the control tower, the questions every agent must answer before launch and the evidence required for continued funding.
Test your readiness before funding another AI agent
Use the ADAPT AI Delivery Readiness Assessment to evaluate how prepared your organisation is to build, run, optimise and govern AI agents. Assess your AI delivery readiness now: https://ap1.hubs.ly/H01w6j30
Key takeaways:
The control tower shows the whole agent fleet
Individual teams see the agents operating inside their own workflows.
They lose sight of the wider system once agents exchange work, drift at different rates and produce different costs.
Vijayan compares the environment to aircraft moving through changing weather.
Air traffic control sees each path, monitors changing conditions and intervenes before routes collide.
An AI control tower serves the same operational purpose.
It brings each agent’s outcomes, costs, drift, failures, handoffs and owner into one view so problems can be traced across the fleet.
The owner then has the information and authority to act when performance changes.
Five questions set the production gate
Every agent entering production must answer five questions:
Three people then sign off on the answers.
The business owner accepts responsibility for value, the technical owner for reliability and the risk or governance owner for controls.
Missing answers expose work that remains unready for production.
Missing signatures expose accountability that has not been assigned.
Two Monday questions guide funding
Every Monday funding decision should begin with two questions: what does a successful outcome cost, and who owns the agent in production?
An initiative with evidence for both belongs in the production portfolio.
Missing evidence places it in the experiment portfolio, with funding tied to a defined question and time limit.
Experiments have value when they produce evidence.
Problems arise when organisations fund them as production systems before establishing cost, reliability and ownership.
Managing AI as a portfolio gives leaders a consistent basis for continuing, redesigning or stopping each investment.
Catch up on Episode 1 to understand the pilot trap, then Episode 2 for the negative unit economics and silent drift that follow in production.
ADAPT Insider’s AI Economics Series - Episode 1: https://ap1.hubs.ly/H01tyyC0
ADAPT Insider’s AI Economics Series - Episode 2: https://ap1.hubs.ly/H01vbY10
Cost per successful outcome determines whether an AI agent creates value or quietly consumes it.
In the second episode of ADAPT Insider’s AI Economics Series, Negative unit economics and the silent drift, Vijayan Seenisamy, Author of The Pilot Trap| Enterprise AI Delivery, joins ADAPT’s Byron Connolly to reveal the costs and performance decline hidden by conventional monitoring.
See what each successful AI outcome really costs
Based on the AI Economics Iceberg in Vijayan Seenisamy’s The Pilot Trap, the AI Economics Calculator reveals the full cost of one agent in three minutes: https://ap1.hubs.ly/H01v0F20
Key takeaways:
Successful outcomes expose the true cost
Token pricing captures only one component of AI economics.
The full cost extends from the initial request to the successful outcome.
Every retry, correction, escalation and manual review adds expense to the transaction.
Vijayan illustrates the gap through a human process costing $1 per completed outcome.
During the pilot, an AI agent appears to complete the same task for 30 cents.
Production introduces failures, rework and oversight. The cost rises to $1.40 per successful outcome.
The pilot presents a 70% saving. Production delivers a 40% cost increase.
Measuring cost per successful outcome gives leaders a direct comparison between an AI agent and the process it replaces.
Volume multiplies weak unit economics
Higher transaction volumes amplify every unresolved weakness in an AI agent.
Probabilistic outputs create retries, escalations and human intervention, with pilot teams manually correcting failures that reported costs exclude.
In production, those failures repeat across thousands of transactions and expenditure grows faster than value.
Measuring unit economics before wider deployment reveals whether scale will improve returns or multiply losses.
Quality drift erodes value over time
Traditional software monitoring tracks availability, latency and errors.
AI agents also require outcome-level monitoring that tests accuracy, relevance and commercial value as source data, customer behaviour and language change.
Quality-over-time monitoring measures each response against a defined business standard.
Combined with cost per successful outcome, it shows whether the agent delivers value after launch.
Catch up on Episode 1 to learn why polished AI pilots fail when they reach production: https://ap1.hubs.ly/H01tyyC0
Enterprise AI pilots often succeed under controlled conditions, then break when messy data, edge cases and dependent systems enter the journey.
In this first episode of ADAPT Insider’s AI Economics Series, Escaping the Pilot Trap and the Demo God Curse, Vijayan Seenisamy, Author of The Pilot Trap | Enterprise Agentic AI Delivery, joins ADAPT’s Byron Connolly to explore why organisations manage probabilistic AI through operating models designed for deterministic software.
The gap surfaces when a successful demonstration reaches production.
AI systems change with data, context and user behaviour, requiring continuous monitoring, clear ownership and business cases that account for reliability across the entire journey.
Key takeaways:
AI requires an operating model built for change
Many enterprise AI experiments run through operating models designed for traditional software.
Code is tested, defects are fixed and the system is expected to behave consistently in production.
AI behaves differently. Its performance shifts as data, context and user behaviour change.
Vijayan describes the distinction as construction versus weather. A building can be completed and signed off. An AI system requires ongoing observation, correction and governance.
Without that operating discipline, a pilot may perform well in controlled conditions while remaining unprepared for production.
A perfect demo proves potential, not resilience
A polished demonstration can create false confidence when its conditions are mistaken for evidence of production readiness.
Demo data is clean, prompts follow a happy path and the environment is arranged to help the agent succeed.
Production introduces messy data, unscripted questions, edge cases and failures in surrounding systems.
The environment has become honest.
Before approving the business case, executives should ask what the test actually covered. A curated demonstration shows potential. Production readiness requires evidence that the system can perform under real operating conditions.
Reliability belongs to the entire chain
Multi-agent journeys expose how quickly reliability falls across connected steps.
Five agents with 92% reliability each produce an end-to-end success rate of about 66% when every agent must perform correctly in sequence.
Each component can pass its own test while the customer journey still fails around one time in three.
Measuring agents individually conceals the reliability of the outcome experienced by the customer.
The business case should price that end-to-end result and assign one person accountability for the entire journey.
Five owners focused on five agents leave no one responsible for whether the chain works.
Control through prohibition cannot keep pace when employees can access AI tools and build solutions themselves.
In this ADAPT Insider episode, Tracey Evans, Group CIO at Virtus Health, examines how technology teams can create safe environments for experimentation while maintaining clear security, regulatory and compliance guardrails.
She also discusses how business ownership strengthens delivery and how new interfaces can improve the experience of legacy systems.
Key takeaways:
Guardrails give experimentation a safe path
The democratisation of AI allows employees to test tools and create solutions without waiting for a technology team to direct every decision.
Tracey’s approach at Virtus Health gives clinicians and business teams controlled environments for learning and experimentation.
Security, regulatory and compliance guardrails establish clear boundaries around what teams can test.
Technology teams work alongside users to understand the problem before introducing a solution.
This positions technology leaders as educators and guardians who help employees apply new tools responsibly.
Safe pathways allow useful ideas to develop while reducing the risk that unmanaged experimentation reaches sensitive data or production systems.
Business ownership strengthens delivery
Technology delivers stronger outcomes when the people closest to the problem help shape the response.
At Virtus Health, each business problem requires a business sponsor and a team that works alongside technology throughout delivery.
This shifts technology away from a back-office order-taking role and connects it directly to frontline priorities.
A digital consent program across 46 clinics demonstrates the model.
Clinical leaders helped define, standardise and refine the process, while the technology team supplied the capability and guardrails required to digitise it.
Virtus Health digitalised 26 consent processes within 12 months.
Clinical stakeholders also became champions for the change because they had shaped the program from problem identification through to production.
Legacy systems can support a modern experience
Legacy ERP, electronic medical record and laboratory systems remain central to operations at Virtus Health.
Replacing every underlying platform would introduce significant cost and operational risk.
Tracey focuses on securing existing databases and building new interfaces over them.
AI-enabled capabilities can then give clinicians and operational teams more useful ways to access the information those systems contain.
This approach modernises the user experience while preserving stable systems of record.
It also gives technology leaders a growing role in educating users about security, data governance, APIs and the limitations of AI-generated solutions.
Clear technical guidance helps teams turn early experimentation into solutions that are safe, scalable and ready for production.
AI pilots can create false confidence.
Controlled data, prompts and supervision often make early results look stronger than the systems that eventually reach production.
Leonard Rawat, Head of Infrastructure, Security and Operations at Harris Farm Markets, explains why infrastructure choices must start with the use case.
He examines what changes as token demand grows, why guardrails belong in the architecture and how internal capability turns experimentation into practical value.
Key takeaways:
Production exposes the weaknesses pilots conceal
Many AI projects perform well in pilots because the data is clean, prompts are controlled and users receive close supervision.
Those conditions disappear when systems enter day-to-day operations.
Leonard says organisations should plan for scale from the outset.
Token consumption can rise quickly as experimentation develops into agentic workflows or automation, increasing pressure on the platform and its operating cost.
The AI platform must align with data location, expected demand and the original business case.
Infrastructure choices determine whether a promising pilot can deliver reliable value in production.
Guardrails have to be part of the architecture
AI infrastructure extends beyond keeping systems available.
It must account for cloud spend, token usage, security exposure and the wider ecosystem surrounding each model.
Leonard argues that clear controls are needed at the infrastructure layer before deployment expands.
Sandboxed environments give teams room to test ideas without exposing production systems, while token optimisation helps contain costs as usage increases.
Embedding these controls into the architecture supports confident scaling.
Without them, risk and expenditure can grow faster than the value created by the technology.
Internal capability turns business context into value
At Harris Farm Markets, employees across marketing, finance, procurement and warehouse functions can use sandboxed environments, training and accessible development tools to build prototypes safely.
Leonard points to practical applications ranging from Shopify workflows to security vendor mapping.
People closest to an operational problem can test a solution in a controlled environment, shortening the distance between an idea and a useful outcome.
Organisations cannot buy their own business context.
They need people who understand the work, supported by the tools and governance required to apply that knowledge.
Most AI initiatives stall when ambition outruns clarity.
Dean Nelson outlines how leaders can reduce that risk by applying first principles, proving value through targeted use cases, then using each result to guide the next investment.
His perspective links practical AI adoption to the infrastructure choices required for sovereignty, sustainability and public trust.
Key takeaways:
Start with one business problem
The strongest AI programs begin with a specific operational constraint rather than a broad transformation mandate.
Dean says many pilots fail because teams select the technology before agreeing on the result the organisation needs.
A first-principles approach brings business, technology and security leaders around one measurable objective.
It also creates a practical sequence: solve the immediate problem, validate the outcome, then expand from evidence.
Dean points to an organisation in Mexico that built a secure private AI environment for field teams.
By focusing on one defined need, it achieved meaningful productivity gains without making enterprise-wide transformation the starting condition.
Build infrastructure that keeps AI under control
AI workloads are changing the physical design of data centres.
GPU-dense environments require new cooling methods, with liquid cooling and closed-loop water systems supporting greater computing density while reducing water use.
The same infrastructure decisions shape sovereignty.
Dean argues that governments and enterprises need locally hosted capacity, private models and Australian-controlled digital capabilities so critical data and services remain within national control.
Sustainable AI therefore depends on architecture that can support intensive workloads without weakening security, operational resilience or accountability.
Public trust determines whether AI can scale
Technical progress alone will not secure the future of AI infrastructure.
Communities increasingly scrutinise the environmental impact of data centres, yet the industry often struggles to explain improvements in renewable energy, efficient cooling and resource use.
Dean calls for open engagement backed by tangible local benefits.
A stronger public narrative must show how infrastructure supports economic activity, improves efficiency and responds to legitimate community concerns.
Inside the organisation, that trust starts with disciplined governance.
Leaders should manage workloads deliberately, use AI to strengthen oversight where appropriate, then measure progress against business outcomes rather than market hype.
For the first time, organisations can measure and buy certain kinds of cognitive work at scale.
Once thinking becomes measurable through tokens, the economics underneath a business start to shift.
Work that once sat inside roles, teams, and service models can now be broken apart, reassessed, and redistributed across people and agents.
In this ADAPT Insider conversation, Mark Pesce, ADAPT Executive Advisor and Co-Founder at Wisely AI, argues that this is why the current moment is bigger than another wave of software.
Businesses are not simply adopting a new tool.
They are being pushed to rethink the cost of work, the design of processes, and the basis of competitive advantage.
Key takeaways:
AI is becoming part of the infrastructure
Mark’s view is that AI will disappear into the background in the same way electricity did.
It will stop being treated as a standalone capability and become embedded across products, workflows, and decisions.
That is why he uses electrification as the comparison.
Businesses that once bolted electricity onto old factory designs were eventually overtaken by those that rebuilt around distributed power.
He sees the same pattern emerging with AI.
The larger gains will go to organisations that redesign around agents rather than leaving AI in a central team or treating it as a side feature.
The real pressure is economic
Mark is less interested in AI as a novelty than in what it does to the economics of cognition.
Once organisations can buy certain kinds of thinking directly, they have to revisit where expensive human effort is still essential, where agents can take on part of that load, and how that changes their margins and operating model.
That is where he sees the real disruption.
Many businesses are still running on assumptions formed before this shift began.
The bigger risk is discovering too late that the business is still priced and organised for a world that no longer exists.
Redesign will matter more than optimisation
Mark is clear that the biggest gains will not come from adding AI to existing workflows and hoping for incremental improvement.
The larger opportunity comes from redesigning those workflows around agents and feedback loops.
He points to software engineering as an example, where agents are already being used to detect issues, fix code, test changes, and redeploy continuously.
In that kind of setup, the work does not disappear, but it changes.
Humans move up a level, from executing repetitive tasks into designing, guiding, and improving the loop itself.
Saving time creates little value when organisations have no deliberate use for the capacity released.
In this ADAPT Insider episode, Dom Price, Work Futurist and Visionary at Be Luminous, examines why conventional productivity metrics reveal too little about the conditions that sustain performance.
He considers how organisations can redirect saved time, protect long-term workforce capability and build the confidence to experiment amid constant change.
Key takeaways:
Saved time requires a deliberate destination
Productivity, sales and GDP remain common measures of performance.
These lagging indicators reveal little about whether people have the environment, tools, trust and space to produce meaningful work.
Dom points to organisations that achieved significant productivity gains, then filled the released capacity with more meetings.
Without a clear next-best use for that time, greater efficiency changed the volume of activity without improving its value.
Leaders can assess the conditions behind performance alongside the final output.
Discretionary effort, curiosity and human connection reveal whether people feel engaged enough to contribute beyond the minimum expected of them.
Automation decisions shape future capability
AI can improve work by augmenting employees and redirecting their attention towards higher-value contributions.
Removing roles without considering the talent pipeline risks leaving organisations without the experience required for future leadership and specialist positions.
Graduate roles illustrate the long-term value at stake.
Early-career employees may initially cost more than they contribute, yet those roles build context, experience and institutional knowledge over time.
Dom points to IKEA’s decision to redeploy employees affected by automation into new roles.
Their existing expertise remained inside the organisation and supported a different source of value.
Workforce planning should account for the capabilities an organisation will require in the future, including the pathways through which employees develop them.
Learning velocity helps organisations adapt
Many organisations were designed around stability, efficiency and five-year planning.
Constant disruption has reduced the value of waiting for certainty before acting.
Dom describes learning velocity as the ability to absorb new information, adjust direction and test ideas without punishing people for changing course.
Leaders can select one manageable opportunity each quarter, test it at low cost and share the result.
A small experiment gives teams evidence they can act on while limiting the consequences of failure.
Leading by example gives employees permission to try, learn and adapt.
Repeated experimentation builds practical knowledge faster than waiting for a perfect business case.
From the publisher's feed