What if the future of Artificial Intelligence wasn’t about ever-larger models or astronomical computing power, but about smarter, more purposeful application? This was the theme running through a recent episode of the Cambridge Executive Business Insights: Rethinking AI podcast, led by Jaideep Prabhu in conversation with Viren Lall, Managing Director of ChangeSchool. They called for a reframing – from viewing AI as a tool for quick fixes to cultivating disciplined, strategic use that compounds value and capability over time.
Beyond the hype: why “Frugal AI” matters
AI occupies a paradoxical space. On the one hand, we marvel at its capacity to generate limitless options, automate routine decisions, and increasingly, to act autonomously in real-world settings. On the other, there’s concern about runaway costs; of computation, of energy, and, less visibly, of human attention and capability.
The notion of ‘frugal AI’, as championed by Viren Lall and explored by Jaideep Prabhu, challenges the traditional narrative. Frugality here is not about scarcity or compromise, but recognising constraints as opportunities for innovation and impact. It asks leaders to shift focus: from how big, fast, and flashy their AI is, to how disciplined and strategically they are allocating their most precious resources, such as their own time, their teams’ attention, and the collective judgment embedded in workflows.
From model size to leader attention: rethinking the frugal constraint
In common discourse, frugal AI often refers to the supply side: modestly sized models, leaner compute, or reduced carbon footprints. But as Viren Lall points out, for leaders in organisations where AI is already ubiquitous, on every desk, woven into every workflow, the constraint is no longer technical, but cognitive. As he puts it, “You can always buy more tokens, but you can’t buy more leaders’ thinking and attention time.”
Just as the revolutionary AI paper “Attention is All You Need” repositioned the importance of focus within technical models, leadership must also shift its own locus of frugality from hardware or budgets, to how decision-makers allocate and protect their attention and judgment in an age of AI abundance.
The AI allocation matrix: deep work vs. shallow hit
Viren offers the AI Allocation Matrix, which is a practical tool mapping how professionals engage with AI along two axes:
- Cognitive demand: how much deep, sustained thinking is required?
- Potential for compounding returns: does the work produce future leverage, i.e., capability and learning that pays back well beyond the immediate task?
Drawing inspiration from Cal Newport’s “Deep Work” and Elizabeth Grace Saunders’ work on time investment, Viren urges leaders to audit their use of AI. Are they merely summoning AI for low-impact, quick wins, what he memorably labels the “candy machine” quadrant, or purposefully applying it to work that, once solved, will pay back in saved time, improved insight, and stronger organisational memory?
The “candy machine” trap
Too often, organisations fall into the ‘candy machine’ mindset: feed an AI tool a prompt, get a sugar hit of output. Draft this email. Summarise that board paper. Polish up a presentation. It feels productive, but as Viren Lall warns, such shallow use rarely results in lasting value or improved capability. “You didn’t learn anything. You didn’t produce any output that was going to generate value over time, but you got a quick result back.”
Escaping the “candy machine”: the AI investment loop
How do those few organisations avoid the trap, and convert time spent with AI into strategic, compounding returns? Lall recommends what he terms the AI investment loop:
- Challenge – before settling on the AI’s first answer, challenge it by asking for alternatives, requesting counter-examples, or interrogating the problem framing. This not only reduces anchoring bias but also forces a broader, more creative exploration.
- Check – calibrate the threshold for quality depending on the risk profile of the task. For a routine email, a cursory check might suffice. For a strategic board paper, more rigorous scrutiny is needed. The burden for this judgment lies squarely with the leader, not with an abstract ‘AI governance’ somewhere upstream.
- Capture– this is theoften-neglected final step: institutionalise what works. When a good result appears, embed the prompt, context, and learning into shared memory. Over time, this creates a unique, context-rich knowledge library that no generic AI model can replace.
Researchers such as Kate Niederhoffer and Jeff Hancock describe this as the “pilot mindset”: staying in the cockpit, leveraging AI’s automation but never relinquishing ultimate human oversight.
The skim tax: the hidden cost of careless AI
Much as the compounding effect of disciplined use of AI builds capability, the inverse is also true. Viren Lall identifies a phenomenon he calls the “skim tax”: the cumulative, often invisible penalty an organisation pays for surface-level use of AI. When individuals pass off AI outputs as their own without scrutiny, organisations drift into mediocrity and risk, with repercussions in two currencies: quality (immediately) and long-term capability (over time).
This “tax” is not just personal, it also ripples downstream. Colleagues lose trust. Organisational judgment atrophies. The chance to coach or apprentice others evaporates. Templates become stale, and teams risk becoming indistinguishable “photocopies of each other”. As Viren notes, “Non-conscious use of AI can really degrade organisational capability at a massive level, if unchecked.”
Scaling up: from pilots to institutional capability
Despite the proliferation of AI pilots, many stall before achieving enterprise-wide adoption. This is not due to technical limits, argues Viren, but a failure of leadership and governance. “From a leadership seat… it’s a leadership problem before it’s a tooling problem.”
The barriers are not in model selection or vendor contracts, but in surfacing explicit decision rights: Who is ultimately accountable for each decision made with AI? What standards of judgment are being applied? Which values are embedding in AI-supported decisions?
As decisions move to machine speed, these become board-level, not merely technical questions. For high-stakes, contestable decisions, Viren Lall recommends the pattern: AI recommends, human decides. For reversible, low-risk choices, humans can frame and AI can execute, with human auditing. But always, a clear line of accountability and a record of human judgment are needed.
Frugal AI as discipline and design principle
Where, then, does frugality sit for leaders? For Viren, it is as much a chosen discipline as an externally imposed constraint. The real limit (and the real opportunity) is in the deliberate focus of energy and attention. As Daniel Holm, a previous podcast guest, summarises: constraint is not punitive, it is generative – it drives innovation.
With abundant AI models at their disposal, leaders must exercise self-imposed limits. The temptation to overproduce, such as generating “ten strategies by lunchtime”, must be countered by a renewed focus on quality convergence: selecting, refining, and executing the options with the best alignment to organisational context and resources.
Competitive advantage in the age of ubiquitous AI
If the underlying models are accessible to all, where does competitive advantage shift? Viren Lall identifies three enduring sources:
- The data flywheel – organisations that systematically capture, challenge, and feed their own experiences back into their workflows create a compounding source of proprietary, contextual knowledge. This becomes a durable moat, which cannot be replicated by off-the-shelf models.
- Organisation design – the design of flows, roles, and routines around AI matters. Teams that integrate AI deeply, coordinate across silos, and embed best practices are better positioned to achieve meaningful, hard-to-copy improvements.
- Judgment compounding – over time, organisations that document, refine, and reuse effective patterns of judgment in their use of AI build a kind of collective intelligence, organisational “muscle memory” that strengthens with each virtuous cycle.
Lessons from the field: manufacturing, legal, and beyond
While headlines often focus on technology giants or elite consulting firms, Viren points to SMEs in manufacturing as strong examples of disciplined AI use. With limited resources but engineering mindsets, these organisations have automated complex quoting, scoping, and even strategic decision-making, not by chasing flashy tools, but by building systematic investment loops that harness their own data and judgment for compounding effect.
By contrast, universities and higher education institutions often remain risk-averse, shackled by concerns over bias and reputational risk. The result, Viren argues, is a missed opportunity for capability building and operational efficiency: a challenge, but also an opportunity for forward-thinking institutions willing to establish robust, disciplined approaches.
AI and the human element: leadership and organisational renewal
As AI matures, it automates ever more routine, predictable work. What remains for leaders is not the easy part, but the difficult, distinctly human aspect: judgment, synthesis, relational engagement, and stewardship of capability. Far from making leadership easier, AI raises the bar for what good leadership looks like.
Leaders who thrive, Viren Lall predicts, will be those who:
- Fight the urge to treat AI as a mere productivity tool, using it instead as a co-pilot for deep, strategic work.
- Protect “the things that do not automate”, that is the human accountability, judgment and context that give organisations their unique edge.
- Institute rituals (such as weekly strategic investment loops) that compound capability, not just deliver outputs.
Practical steps: the weekly AI investment loop
For those seeking to catalyse this shift, the most actionable step is simple but profound: each week, select one genuinely strategic task and run a full AI investment loop: challenge, check, capture, aloud, with a colleague or team. Make the thinking and learning visible, not just the answer. Over time, these loops accumulate, embedding disciplined, context-rich processes at all organisational levels.
Reimagining AI strategy from the ground up
AI’s promise is not in its novelty or its raw power, but in how it is woven into the lived reality of teams and leaders. To move beyond the “candy machine”, organisations must cultivate the discipline to use AI for deep, compounding work. Frugal AI is not a constraint to bemoan, but a design principle: it shapes how we invest our scarcest resources – time, attention, and judgment – for the greatest, most lasting impact.
To dig deeper into these insights and more, tune into the full episode of Cambridge Executive Business Insights: Rethinking AI, wherever you get your podcasts. Discover how disciplined, human-centred strategies can unlock AI’s full, sustainable potential.
