By Keshavan Thiagaraja
If AI becomes dramatically better at writing software, what should business leaders change today? My answer is to measure how it improves a complete business process, including the work needed to check and maintain its output.
Could AI add US$20 to US$30 trillion to annual global economic output? Elon Musk has advanced that rough estimate, alongside predictions that AI could handle almost any digital task and reach Stockfish level coding ability. These claims have attracted attention because they suggest a major shift in the economics of knowledge work.
These are ambitious forecasts, not established outcomes. Business leaders should distinguish the possibility of major productivity gains from proof that AI can reliably deliver every digital task. Investment decisions need evidence from the work an organisation actually performs.
What the Stockfish comparison tells us
Stockfish is an open source chess engine and one of the strongest in the world. The comparison illustrates a real possibility: machines can outperform people in a demanding intellectual task. However, the analogy alone cannot establish when AI will outperform every software engineer.
Chess has fixed rules, a visible board and a clear objective. Enterprise software involves incomplete requirements, changing policies, legacy systems and competing stakeholder priorities. A programme can pass its tests while solving the wrong business problem.
Generating code is one part of engineering. Deciding what to build, controlling access, checking security, deploying safely and supporting users remain part of the outcome. Even highly capable AI needs a process that can detect mistakes and assign responsibility.
Understanding the economic claim
A 20 to 30 per cent increase in the level of annual economic output is different from 20 to 30 per cent growth every year. A claim of this scale needs explicit assumptions about adoption, reliability, investment and the time required to realise gains. Pairing those percentages with US$20 to US$30 trillion implies a rough starting economy of US$100 trillion.
For perspective, Goldman Sachs published a 2023 scenario in which generative AI could raise global GDP by about 7 per cent, or almost US$7 trillion, and increase productivity growth by 1.5 percentage points over a ten year period. That was a conditional forecast, not a measured result or a ceiling on future gains. It does not independently validate Musk’s larger estimate.
At company level, a useful calculation is simpler: net value equals the value of additional useful output and realised savings, less AI costs, integration, review, rework and losses from errors. Time saved becomes financial value only when the organisation can use it productively.
The advantages and the costs
Speed versus verification. AI could shorten drafting, coding and analysis. Leaders should measure the time needed to check the result alongside the time saved producing it.
Lower production cost versus new operating cost. Creating a first version may become cheaper. Model usage, integration, monitoring and correction still consume money and staff time.
Wider access versus dependence. Smaller teams could undertake work previously beyond their resources. Dependence on a single provider could make prices, availability and model changes harder to manage.
Greater capacity versus workforce disruption. Staff could spend more time on customer needs and difficult decisions. Businesses also need to plan for changing roles, retraining and the development of junior employees whose routine learning tasks may be automated.
A practical enterprise example
Consider a hypothetical warehouse business building an order tracking service. AI could draft the interface, propose database queries and generate test cases. The team would still need to confirm which customers can view each order, how stock discrepancies are handled and whether the service works with existing systems.
The meaningful comparison is cost per accepted release and performance in use. Measure delivery time, defects, support requests and customer outcomes. A faster first draft that creates more production incidents may reduce overall value.
The same principle applies to customer service. Automating straightforward requests could improve response times. Unclear cases need a reliable path to a person, with the relevant context preserved.
What CEOs and CTOs should do
CEOs should choose one process with an identifiable cost and a clear customer outcome. Establish a baseline, run a limited pilot and agree in advance what improvement would justify expansion. Include implementation costs and changes to staff responsibilities.
CTOs should evaluate the proposed model on representative work from their own systems. Check correctness, security, privacy, recovery from failure and compatibility with existing services. Separate permission to suggest a change from permission to execute it.
For consequential actions, define the approval owner and keep an audit trail. Retain the ability to reverse changes, restrict access and suspend automation when performance deteriorates.
Review supplier dependence and train staff to challenge outputs. Expand deployment when measured benefits remain positive after review and support costs. A global economic prediction should inform scenario planning, while local evidence should guide investment.
The leadership decision
AI may substantially change the economics of digital work. The scale and timing remain uncertain. My advice is to build the capacity to evaluate and adopt it responsibly now, then increase investment as the evidence strengthens.
Which process in your organisation would you test first, and what evidence would convince you that AI had created lasting business value?
Think mathematically. Engineer intelligently. Lead strategically. Build responsibly. Scale globally. Create real business value through artificial intelligence.
#ArtificialIntelligence #AIStrategy #EnterpriseAI #SoftwareEngineering #Productivity #BusinessLeadership
Originally published on LinkedIn by Keshavan Thiagaraja. Read the original on LinkedIn.
