The AI Productivity Paradox
AI does not create productivity.
There, I said it.
A new report from the Economist, an academic’s prediction and some wider discussions got me back on this topic this week.
Productivity has always been a paradox.
You only have to look at the work of economist and Nobel laureate Robert Solow, many years ago, who highlighted the contradiction that despite rapid advancements and heavy investment in technology, productivity growth in developed economies remained stagnant.
Now we have all read the claims by various companies reporting job cuts “due to the advances in AI”. However, the evidence of “re-hiring” and the lightest scrutiny of the facts vs the positive stock price impact of an announcement that sounds more optimistic than “we need to cut costs”, tend to undermine many of these claims. When validating these announcements, it always seems good advice to “follow the money” . . .
Talking to the WSJ Leadership Institute, the Stanford economist Erik Brynjolfsson forecasts that “Productivity will be Much Higher Due to AI” and argues that the “J Curve” of AI adoption is turning upward. Brynjolfsson estimates 2025 productivity growth at 2.7% – double the previous decade’s average.
Impressive. But still a prediction.
The challenge with productivity is that it is a collective economic measure, not one of individual, personal experience. It is the ratio of inputs (labor, capital, materials, cost) to outputs (goods, services, revenue).
In conversation, we often confuse productivity with the amount of effort we individually apply to a specific task, how many tasks we perform per day or even the number of hours we work per day.
A new report from Harvard Business review (HBR) this week explores a compelling argument, “AI Doesn’t Reduce Work—It Intensifies It”. The authors report;
- “One of the promises of AI is that it can reduce workloads so employees can focus more on higher-value and more engaging tasks.”
- “But according to new research, employees worked at a faster pace, took on a broader scope of tasks, and extended work into more hours of the day, often without being asked to do so.“
- The interesting question is that even if it was a good thing for people to work harder and longer due to “AI pressure”, does this actually generate improved economic productivity, in terms of revenue growth and productivity?
- The jury is out.
Following this theme, Sven van de Kamp summarised this impact of Generative AI, specifically, with keen precision;
- Using (Generative) AI well requires two things most people skip.
- 𝗙𝗶𝗿𝘀𝘁, you need to know how to prompt. Context, framing, specificity. Table stakes . . .
- 𝗦𝗲𝗰𝗼𝗻𝗱, and this is where it really breaks down, you need enough knowledge to evaluate what comes back. AI output reads confidently. But if you can’t tell the difference between a relevant assertion and a hallucinated one, you’re not gaining clarity. 𝗬𝗼𝘂’𝗿𝗲 𝗴𝗮𝗶𝗻𝗶𝗻𝗴 𝗻𝗼𝗶𝘀𝗲.
- If you can’t (or don’t) evaluate the output, you’re not saving time. 𝗬𝗼𝘂’𝗿𝗲 𝗰𝗿𝗲𝗮𝘁𝗶𝗻𝗴 𝗮 𝘀𝗲𝗰𝗼𝗻𝗱 𝗽𝗿𝗼𝗷𝗲𝗰𝘁.
In this AI age, we hear a number of repeated assertions related to productivity, many of which also apply to classic deterministic technology. Some are plain wrong, such as “AI automates the mundane, repetitive tasks” (that is the role of traditional deterministic, rather than probabilistic, software).
AI is probabilistic and should be targeted at tasks where that is a desirable approach, forecasting (which has always been probabilistic, even when didn’t know it! 😉), for example.
The Economist report “Agents of Change – The Rise of the Autonomous AI Enterprise” aims to explore the rise of Agentic AI but, in common with so many reports, conflates AI in general (Machine Learning, Analytical AI, Generative AI etc) with Agents , Agentic AI and traditional deterministic digital automation.
- The report suggests that “Finance teams have been early adopters, tracking cost per invoice, days to close and forecast accuracy, which help show whether agents earn their keep.”
- Interesting to debate whether “cost per invoice” is a direct measure of AI impact or a correlation to something else . . .
- The report suggests that the “predominance of high-volume work like transaction monitoring suggest a valuable supportive role for agents”. Hmmmm – Feel free to disagree with me, but I contend that this is not the sweet spot of AI in general, or Agentic AI specifically.
- The upward trend graphs represent, on examination, “industry references to GenAI and Agentic AI, 2022 and 2024” rather than specific experience and delivery of superior business outcomes.
One thing has become clear, as if we shouldn’t have known already, that AI alone doesn’t create productivity.
We already knew that technology alone doesn’t create productivity, so why should a new generation of tech do so?
The business outcome and problem to which AI (or other technology) is targeted is the KEY. This is where failure typically starts, in the “silo”. Rethinking how best the business outcome can be optimised (revenue, profitability, customer success et al), enhancing the end-to-end process that delivers that outcome, adapting to human behaviour (internally and externally) and ONLY THEN working out how the AI (or any other tech) is designed, embedded, and trusted in real workflows in that end-to-end process is the sequence that matters.
- “Start with the end in mind”.
- The business outcome.
- Delivered through an end-to-end process.
- Diagnosed with experience and data, including the “Blind Spots”
- Aligned, simplified, enhanced.
- Only then applying the right mix of the amazing technologies we have available to the points of maximum effect.
This is how true productivity is improved.
You can read Sven van de Kamp’s incisive post on the productivity impact of Generative AI here . . .
The AI Productivity Paradox
Thanks for reading . . . .