1️⃣ Watch the work before designing the training MIT SMR studied 10 European manufacturers for three years. The useful skills created by new technology rarely matched forecasts; workers developed them while adapting tools to real work.
💡 Why it matters Put managers close enough to the work to spot new practices, name them and change the process. A course catalogue will arrive late.
☕ Coffee talk Who can actually see the workaround on the factory floor before HR turns it into a competency?
2️⃣ Redesign the work before cutting the role HBR argues that early AI-driven layoffs have often missed expected returns and removed organizational knowledge. Goldman Sachs estimates AI added only 0.1 percentage points to US unemployment over the past year.
💡 Why it matters Map tasks, judgment and handoffs before changing headcount. A smaller org chart does not prove the work now runs better.
☕ Coffee talk Which “AI saving” disappears the first time the team needs the person who knew the exceptions?
3️⃣ Run two real bets when prediction is cheap Microsoft CoreAI EVP Jay Parikh would rather fund two competing approaches in parallel than pretend the winning architecture can be forecast early.
💡 Why it matters Define the learning each bet must produce, cap the spend and set a decision date. Parallel work is useful only if it buys evidence.
☕ Coffee talk Would your second bet survive budget review, or is it there to make the first one look tested?