I follow several podcasts about AI and automation. They always return to data entry, research and coding. While future machines will be able to do our jobs, when pressed about the present, pundits fall back on recurring examples.

There is a sense of frustration in the Silicon Valley mindset. It comes from the 17-hour days, the all-or-nothing competition and the desire to disrupt. If you have these beautiful new tools why would you not use them. The power of AI lies in the reinvention of business.

The rest of the world is more circumspect. Advanced knowledge is required to get the most out of the latest models. It’s tough to prompt LLMs to provide PhD level research if you don’t have a PhD. People know that not all innovations show up in higher profit margins, other than for the tech companies.

The tech optimists tell us that AI will reinvent businesses by figuring out how to do things better. There are reasons why this may be slow to take-off. Investment bankers tell us they can automate 90% of corporate finance work. Yet the last 10% is where human judgement provides the value. Lawyers say something similar. Value lies in the liability of an opinion, after all the research and documentation has been done.

Those who argue that productivity will explode to an unprecedented 10% or more, assume AI replaces human judgement. Those who expect automation of existing processes, while reserving human judgement, expect a number closer to 2% a year. Even that would represent a golden age of economic growth.

Distilling economy-wide questions of productivity to the firm level raises questions for business owners. Where is scarcity a bottleneck in your business? Is there a way to introduce abundance into the process while maintaining your margins? The answers are revealing.

Azeem Azhar identifies three inputs to human progress. These are energy, intelligence and biology. All three have been through a fundamental change. Where once they were scarce, now they may be built and improved.

Renewables mean energy becomes technology. We no longer have to burn resources to generate power. Energy is susceptible to Wright’s Law, whereby costs fall by a fifth for every doubling in production.

Intelligence became scalable with the advent of transformer architecture. This enabled sequences to run in parallel using self-attention. Intelligence became a function of data and compute.

Perhaps most remarkable is the change in biology. The ability to read and edit the human genome gave rise to CRISPR and mRNA vaccines. The cost of analysing individual genomes has fallen ten thousand times over 20 years.

Azhar then relates these changes to three constructs of the modern world. These are jobs, credentials and expertise. All three are reaching their sell-by date.

Jobs solve the question of how to organise production and distribute resources. Agentic AI unbundles them into component tasks. Credentials solve for trust, which we outsource to educational institutions. Nowadays it is cheap and easy to have candidates prove their worth, by making marketing videos or software applications. Expertise is the bottleneck resulting from scarcity of information. AI makes knowledge limitless.

Business owners intuitively break work down into tasks. Their first instinct is to ask can I do this. The second is to outsource it. Only large companies have the luxury of hiring their own accountancy, legal or marketing departments. The initial adoption of AI is about giving owners more control over outsourced work.

The next phase is to ask where can I pour abundance into my workflows. This may be best explained through examples of businesses we are working with at MSBC.

Design is a key element of construction engineering, while payment is for completion. AI tools make it easier to design environments, threatening a pinch point that creates value for the engineer. Rather than defend this niche, an owner builds a tool allowing people to prompt and price their own designs. He reaches a wider audience in new fields, while retaining control of the blueprints until contracts are exchanged.

A school has knowledge about students’ schedules, the curriculum and entry requirements into higher education. This creates a stream of repetitive questions from parents who expect personalised responses. An AI tool tapping public and private information delivers this and allows teachers to focus on providing results for parents.

An insurance company houses expert knowledge about underwriting of esoteric risk. This is stored in a history of handwritten notes using industry jargon. The company deploys the latest image recognition technology and captures, shares and reuses its inherent advantage.

The value of a firm is more than just execution. It lives in sequencing, judgement and accountability. The final 10% is where margins are defended and trust is earned. Automate it too early and you flatten what makes the business valuable.

A practical starting point is to split work into three types. What you do, what you know and what you decide under uncertainty. AI is already strong in the first two. That is where speed, cost and consistency generate returns.

The third deserves restraint. Use AI to prepare inputs, options and challenge assumptions. Keep humans responsible for the call. If a mistake would be expensive, reputational or irreversible, that decision should not be automated.

The firms that benefit most from AI will not be the fastest to replace people. They will be the ones that pour abundance into their workflows, while protecting the pinch points where judgement provides value.