Category:
AI Strategy
Duration:
8 minute read
Introduction: Beyond AGI
Discussions about Artificial General Intelligence (AGI) dominate public discourse on AI. While the notion of machines matching human capabilities across all domains captures the imagination, it distracts from the more immediate transformation already reshaping our workplaces.
This essay proposes a practical mental model to help knowledge workers and business leaders survive this transformation: Replacement Level Intelligence (RLI).
"
RLI refers to the threshold when a particular task or subtask performed by a knowledge worker can be executed by AI at a level of capability and cost sufficient to replace the human in performing that specific function.
"
This isn't a theoretical future milestone. RLI is observable today, advancing incrementally across numerous business functions. It is what will matter most for companies navigating the "AI Transition" - the period we are living in now during which businesses and knowledge workers progressively increase the amount of AI technology they use.
If we consider OpenAI's definition of AGI - "highly autonomous systems that outperform humans at most economically valuable work" - then the development path toward AGI will be paved with RLI, consisting of numerous incremental instances of task-by-task replacement. Long before AGI arrives, companies implementing RLI will gain advantages over those that don't. And workers who effectively leverage RLI will outperform those who fail to adapt.
This inaugural essay for Proof To Production - you might think of it as a position paper - establishes the blog’s central thesis: the competitive battle will be won or lost on the journey to AGI, not when it arrives.
It also attempts to provide a playbook - which will no doubt evolve over time - for establishing an AI-first culture, that enables organisations to secure a critical first-mover advantage as the AI Transition unfolds.
Historical Patterns in Technological Disruption
The business landscape provides numerous examples of once-dominant companies that failed to adapt to technological shifts.
Kodak, which developed early digital camera technology but continued prioritising film products, eventually filed for bankruptcy as the market transitioned decisively toward digital imaging. Blockbuster declined an opportunity to acquire Netflix for $50 million, maintaining focus on physical stores while video distribution moved to streaming services. Nokia dominated mobile phones but failed to transition effectively to smartphones. Borders expanded physical bookstores while Amazon pioneered e-commerce.
These examples, while simplified, illustrate a recurring pattern: market leaders can fail to recognise or respond effectively to fundamental technological shifts until their competitive position becomes compromised to an unsalvageable extent.
In technological revolutions, the relative advantage gained by early adopters is eroded as the wider market implements similar capabilities. The same will apply in the AI Transition as each unit of RLI becomes widely adopted. The bulk of the upside will go to the early movers before the advantage is competed away. Likewise, for the laggards, the longer they take to catch up, the less impactful catching up will be, although the cost of doing so will remain.
The competitive dynamics of the AI Transition will likely follow similar patterns to previous technological revolutions, but what potentially distinguishes it is the pace of capability improvements and the breadth of implementation possibilities. The digital revolution disrupted certain industries - like music, film and print media - more than others. The AI Transition (which will increasingly include embodied AI) will disrupt all job functions and processes in every industry. This may amplify competitive dynamics and increase the scope of opportunities for early adopters.
Taking all of this into account, it is reasonable to infer that companies which integrate RLI at scale into their operations early on, will replace competitors who significantly delay adoption. The AI Transition will be as much about competitive advantage as it is about survival.
A Framework for Analysis: Companies as Data Processing Systems
DeepMind Co-founder and current CEO of Microsoft AI, Mustufa Suleyman, wrote in his recent book, The Coming Wave, “These tools will only temporarily augment human intelligence. They will make us smarter and more efficient for a time, and will unlock enormous amounts of economic growth, but they are fundamentally labour replacing.”
"AI is coming for our jobs", another favourite for dinner party debates. In the meantime, businesses need a way to understand how this transformation is unfolding in the here and now.
A useful mental framing is to conceptualise a company as an interconnected system of nodes that process data. The comparative efficiency and effectiveness of this processing, relative to competitors, constitutes a significant portion of a company’s competitive advantage. This system typically contains three types of nodes, which act in concert to process data in different ways:
Human intelligence nodes
These represent tasks or subtasks performed by humans within the overall system. While indispensable today, these nodes have inherent limitations in cost, scalability and reliability.
Legacy deterministic software nodes
These consist of conventional, rule-based software and integrations, like SaaS applications, APIs and traditional databases. These have historically automated routine tasks but lack the nuanced adaptability of human cognition.
Artificial intelligence nodes
AI nodes are dynamic systems employing machine learning, large language models and other AI techniques. Once RLI-ready for a particular function within the network, they are better, faster and cheaper than the human intelligence node they replace.
Currently, AI nodes represent a small minority of the overall system in most organisations, but their proportion will increase over time as each human task or subtask succumbs to RLI. The AI Transition will be characterised by the progressive replacement of human intelligence nodes with artificial intelligence nodes.
This replacement pattern emerges partly because the fuzzy, probabilistic nature of AI is better suited to replacing certain human cognitive functions. And partly because, as the jagged frontier advances, developing an AI solution - even for structured tasks - can be more cost effective than developing a traditional software equivalent.
To reiterate the key insight behind RLI - human intelligence nodes become candidates for replacement by AI nodes when the capability frontier of AI advances to the point where Replacement Level Intelligence becomes available to perform the specific function at comparable or superior performance and economically viable cost.
This analysis suggests two conclusions:
Organisations with greater proportions of AI nodes within their overall architecture will likely achieve higher efficiency in data processing (translating to economic productivity)
Organisations that create conditions enabling rapid identification and implementation of RLI opportunities will likely accumulate competitive advantages more quickly
Moreover, this advantage will likely be self-reinforcing. As organisations increase efficiency through RLI adoption, they generate additional resources for further AI investment, competitive pricing, increased marketing, or other strategic initiatives.
Competitors who adopt RLI more slowly may experience the opposite effect, with disadvantages compounding as they face reduced pricing power and diminished resources to close the capability gap.
The Current Pace of Development
Let's assume that the six leading Western frontier and open source model developers - OpenAI, Google DeepMind, Anthropic, Meta, Mistral AI, and xAI - collectively release major model updates approximately every six months. This creates an environment where new capabilities emerge monthly. If cost structures change at least once during any 6-month period after release, this represents a shift in either capability or cost every fortnight.
This is without considering the growing number of independent labs, legacy technology companies and academic institutions pursuing AI research, trends in distillation (the process of transferring capabilities from a large model to a smaller more cost effective one) and fine tuning (post-training using a targeted dataset to improve a model’s capabilities for a specialised use case).
While previous technological revolutions unfolded over decades (industrial revolution) or years (digital revolution), certain aspects of AI advancement appear to be occurring on significantly compressed timescales of months or even weeks.
The jagged frontier of AI capability advances fitfully, but relentlessly, creating an environment where opportunities for competitive advantage continually emerge.
Creating Conditions for Effective RLI Adoption
Given these dynamics, how might organisations position themselves to win in the AI Transition? The evidence suggests creating a systematic approach to identifying and implementing RLI opportunities. To compete effectively in the AI Transition, knowledge workers will need an AI-first mindset and the companies in which they work will require an AI-first culture. The following are some possible practical guidelines to achieve that.
In the spirit of Proof to Production’s evidence-based approach, and given how early we are on the AI adoption curve, these recommendations remain unproven. They are best guesses - albeit based on the author’s own real-world experience implementing RLI and anecdotal insights from others exploring the jagged frontier. As such, they should be seen merely as a starting point for further discussion. Debate and contribution from the Proof to Production community should test, build upon, or even refute them, as the AI Transition unfolds.
1. Executive Commitment with Distributed Implementation
Executive leadership must prioritise AI transformation across all functional areas. However, successful implementation will require both top-down commitment and bottom-up execution.
Knowledge workers require direct access to appropriate AI models and tools (including platforms that enable non-technical users to develop and test new AI workflows). This decentralised approach recognises that only individual contributors who possess the detailed understanding of their specific tasks have the contextual knowledge necessary to identify the highest-value RLI-ready opportunities and to validate and implement them effectively.
This suggests that centralised implementation exclusively through IT departments or executive mandates is likely insufficient.
2. Strategic Investment in Capability Development
Organisations must allocate resources for both AI tools and capability development among knowledge workers. This includes training programs to develop understanding of how to effectively identify and implement RLI opportunities.
This is more than merely purchasing an enterprise AI subscription; it requires ensuring workers understand how to leverage these tools within their specific context. The return on these investments has the potential to increase over time as more RLI-ready human intelligence nodes are augmented or replaced with more efficient AI-enabled processes.
3. Create Recognition Systems for Innovation
Establishing mechanisms to recognise and reward knowledge workers who effectively implement RLI solutions creates visibility for successful approaches and encourages broader adoption.
These "AI Champions" can serve as practical examples and mentors for others in the organisation. Cultural transformation around technology adoption typically occurs gradually and benefits from visible examples of successful implementation. Identifying early adopters who are already implementing RLI successfully for their own tasks provides an opportunity to elevate them as exemplars for others.
4. Implement Systematic Evaluation Processes
Organisations will benefit from establishing structured approaches to continuously evaluate potential RLI opportunities as technology capabilities evolve.
The frontier of AI capability continually advances along dimensions of both capability and cost. A task that doesn't meet the RLI threshold today may become viable with subsequent model improvements or cost reductions. A systematic approach to regularly reassessing opportunities in the hands of an empowered workforce versed in agile innovation methodologies will ensure timely implementation as capabilities evolve.
5. Evolve Information Technology Functions
The responsibilities and capabilities of IT and risk management functions require significant evolution. Beyond traditional responsibilities for connectivity, security, and basic technology services, these functions must shift to enabling an AI-first workforce.
This evolution includes:
a) Providing access to diverse AI capabilities: A Microsoft Copilot subscription does not an AI-first organisation maketh. To innovate effectively, AI-first knowledge workers require access to a diverse range of AI models, preferably accessible through a ‘model agnostic’ no-code / low-code enterprise-grade platform with agentic features and a manageable learning curve for non-technical users. Restricting model options unnecessarily constrains the identification of viable RLI opportunities, while technical friction constrains implementation.
b) Enabling secure data access: Effective AI implementation requires appropriate access to relevant organisational data. A human intelligence node can only be effectively augmented or replaced by an AI intelligence node if the latter has secure, permission-aware access to the same information context across all relevant data silos. Cyber security teams can either be valued enablers of, or a bottleneck for, innovation.
Organisations must recognise that this transformation of technical functions may require additional investment in staffing, skill development, and new technological capabilities.
Conclusion: Practical Implications of the RLI Framework
The AI Transition is underway. Given that technology adoption tends to lag behind technology capability, the observable impacts across industries and functions - although small now - will continue to grow, thus increasing adoption urgency over time. This means it’s probably not too late to get ahead - be that of a colleague on the desk next to you, or a company with which you compete. But, this will hold true for only so long.
Replacement Level Intelligence will continue to reach viability for an expanding set of knowledge work tasks as capability frontiers advance and the cost of those capabilities continues to decrease. Proactive organisations that systematically identify and implement RLI will develop cumulative advantages over those that respond reactively to competitive pressure.
This doesn't suggest the wholesale replacement of humans with AI in the foreseeable future. Instead, the AI Transition will play out as an incremental task-by-task replacement by AI able to perform at comparable or superior capability at economical cost - which, taken in aggregate, will result in progressively more efficient processes and competitive operations.
Organisations will inevitably incorporate AI capabilities; the relevant question is whether they will do so with sufficient speed and effectiveness to remain competitive during a period of accelerated technological change.
The time for developing these capabilities is now when the rewards are greatest; not when they become necessary for survival.