Beyond Cost-Cutting: AI Is a Strategy Shift, Not a Headcount Trick

A friend sent me a viral claim: “IBM fired 8,000 for AI – then rehired 8,000 when AI failed.” I looked everywhere for a primary source. There wasn’t one. This headline was just hype and totally misleading. (note if you cannot find proof – don’t believe) The verified story I found is more nuanced – IBM projected that about 7,800 back-office roles could be automated over roughly five years (mostly via attrition) and later said AI agents had replaced the work of “a couple hundred” human-resources roles while hiring grew in engineering, sales, and marketing. That’s a work-mix shift, not a morality play.

The bigger point – and the one leaders miss when they view artificial intelligence (AI) only through a cost-cutting lens – is that the current form of AI behaves like a general-purpose technology: it changes what organizations can do, not just what they spend. History says the real gains show up after you redesign work around the new capability. Paul David’s classic analysis of electrification and the “productivity paradox” is still the best explanation of that adoption curve.

My workflow changed first

This isn’t theoretical. My operating model is different today (and will continue to evolve as the technology evolves):

Research and learning. AI assistants help map a field, surface primary sources, and draft comparisons, create briefs, generate podcasts and much more – I verify the facts with receipts. The time from question to verifiable answer has collapsed.

Purpose-built software, fast. Instead of waiting for a vendor’s one-size-fits-all module, I am spinning up small and large tools tuned to my use cases – company screeners, source-linked research checkers, and content/social ops helpers with audit trails.

Parallel workstreams. Agents handle scaffolding (summaries, drafts reviews of companies, checklists, data pulls), so multiple projects run in parallel without quality slipping.

No more waiting. I’m no longer constrained by needing help to get started; I can move as fast as I can learn and implement. It’s liberating. You still must put in the time and effort – but you can move forward without the drag.

Still early, accelerating curve. Every week the build-deploy loop gets easier as models, memory layers, and tool integrations improve.

That experience is the point: AI isn’t just a line item for cost savings; for me it’s operating leverage.

The cost-cutting trap

If your AI plan begins and ends with headcount reduction, you’re leaving huge value on the table. Electrification didn’t pay off until factories rewired layouts; the internet didn’t pay off until companies rewired distribution and support. Same here. Even the bank-teller analogy holds: automated teller machines (ATMs) cut per-branch costs and shifted teller work toward sales and service rather than eliminating it one-for-one.

Fresh evidence that lifts this out of anecdote: an NBER study of task-level exposure to AI inside firms finds two knobs drive labor outcomes –

higher average exposure of an occupation’s tasks lowers that occupation’s within-firm share, but

higher concentration of exposure in a few tasks raises the share as people reallocate to complementary work.

On strategy and growth (in ‘from cost center to strategy engine’): “Firms that actually adopt AI grow faster – think +9-10 points in revenue, +7-8 points in Total Factor Productivity, and more hiring over five years. The value shows up when you redesign work and scale, not when you freeze it in place.” (This is a huge opportunity for you to drive your company up and to the right)

What the evidence says (without the hype)

Frontline knowledge work. Support agents with an AI assistant: ~14% average lift in issues resolved per hour, with the biggest gains for novices. That’s capacity creation and quality lift, not just labor subtraction.

Software development. Developers with Copilot: ~56% faster on a standardized task – cycle-time compression you can bank on if you redesign code review, testing, and release to capture it.

Reality check on “full automation.” When firms skip the human fallback, quality breaks: McDonald’s ended its IBM voice drive-thru pilot and is reassessing; Klarna rebalanced to guarantee a human option next to its chatbot. These are design lessons, not indictments of AI.

Risk lens. If a chatbot speaks for your brand, you own the output. The Air Canada case (see reference below) made that clear – build reviewability from day one: show sources, expose uncertainty, and make escalation cheap and fast.

From cost center to strategy engine

Re-architect work, not just roles. Put AI where it changes cycle time, error rates, and decision latency – the levers that move revenue and resilience. Start where the evidence is strongest: support (+14%) and software (~56% faster).

Design human-in-the-loop by default. Let AI handle routine steps; route ambiguity, empathy, and exceptions to people. Measure the escalation path (the McDonald’s/Klarna lesson).

Install a trust layer. Receipts (provenance/citations), Confidence (uncertainty bands), Override (fast human control), Audit (logs for accountability). This manages legal and brand risk (see Air Canada).

Fund the measurement. Track First-Contact Resolution, Customer Satisfaction, defect/leakage, and “proof-seconds” (the time to a verifiable answer with sources). If these don’t move, your AI isn’t strategic yet.

Zoom out: countries are playing offense

While some firms fixate on cost takeout, nations are investing in capability – talent, compute, deployment, and trust. Singapore’s National AI Strategy 2.0 is explicit about raising competitiveness across finance, healthcare, logistics, and public service. The United States and United Kingdom are building shared testing capacity between their AI Safety Institutes to accelerate safe adoption. That’s strategy, not staffing math. If your country isn’t thinking like this – be very concerned. AI will change the dynamics of what we think of as the domain of nation states.

Back to the headline

Judging AI by a single number – “jobs cut” – is the wrong scoreboard. The real and verified IBM story is mix-shift and redeployment, not a neat “fired-then-rehired” parable. If you limit AI to cost takeout, you’ll underinvest in the redesigns that create durable advantage.

The questions that matter:

How much faster did we go from question to verified answer?

How much safer did decisions become with receipts and oversight?

How much better is customer experience with a human-in-the-loop?

And, like my own workflow, how much more parallel can your organization run without the drag?

Before joining a fund – I ran a strategy consulting firm for many years and worked with dozens of large multinationals. The companies that will win won’t treat artificial intelligence as a budget exercise – they will leverage the opportunity. They will redesign work around the new capability, keep humans in the loop, and measure what matters: faster time to a verified answer, fewer defects, better customer outcomes. Companies that do that consistently will see that AI stops being a headline about headcount and becomes a durable advantage.

References

Brynjolfsson, E., Li, D., Raymond, L. “Generative AI at Work.” NBER Working Paper 31161 (2023).

Microsoft Research. “The Impact of AI on Developer Productivity: Evidence from GitHub Copilot.” (2023).

Associated Press. “McDonald’s ends AI drive-thru test with IBM.” (June 18, 2024).

Customer Experience Dive. “Klarna reinvests human talent in customer service.” (May 9, 2025).

American Bar Association. “Tribunal confirms companies remain liable for information provided by AI chatbot.” (Feb 2024).

David, P. A. “The Dynamo and the Computer: An Historical Perspective on the Modern Productivity Paradox.” AER (1990).

McKinsey. “The economic potential of generative AI: The next productivity frontier.” (2023).

Government of Singapore. National AI Strategy 2.0. (2023).

Wall Street Journal. “IBM CEO Says AI Has Replaced Hundreds of Workers but Created New Programming, Sales Jobs.” (May 6, 2025).

Reuters. “IBM to pause hiring as it plans to replace 7,800 jobs with AI.” (May 1, 2023).

Bloomberg. “IBM to Pause Hiring for Back-Office Jobs That AI Could Kill.” (May 1, 2023).

McKinsey (PDF). “The economic potential of generative AI.” (2023).

Hampole, M., Papanikolaou, D., Schmidt, L. D. W., Seegmiller, B. “Artificial Intelligence and the Labor Market.” NBER Working Paper 33509 (2025).