One curve, three numbers and one tension, in five minutes. Every figure is checked against the source named alongside it; forecasts and opinions are kept separate from observed data.
Editorial clarification r5 · 8 August 2026: the issue now connects its analysis to precise Practicum worksheets. Facts, sources and the overall conclusion remain unchanged.
What actually changed
The cost of getting an AI result at comparable quality is falling fast. According to Epoch AI, the median rate of price decline was about 50-fold per year, with a wide range across tasks, roughly 9× to 900×.
What cost a ruble a year ago may now cost pennies. At this point, "intelligence became cheap" is a measurable curve rather than a marketing metaphor.
What became cheaper, and what became more valuable
Drafts, analyses, code and other AI outputs are getting cheaper to produce. But an outcome, a decision, adoption, profit, trust or a risk removed, does not appear automatically.
The issue's conclusion: cheap output is not a cheap outcome. As "doing" becomes easier, choosing the direction, checking the work and owning the result become more visible sources of value.
Three signals to verify
- 57% of US work hours are technically automatable: 44% through AI agents and 13% through robots. This is technical potential, not a forecast of layoffs. Source: McKinsey Global Institute, "Agents, Robots, and Us", 2025. Choose one frequent, narrow workflow, map its agent, human and robot steps, and automate only a step that already has a quality criterion and an outcome owner.
- 170 million roles may be created and 92 million displaced by 2030: a net balance of +78 million, with substantial reskilling required. Source: World Economic Forum, Future of Jobs Report 2025. Do not make a career or staffing forecast from one global figure; map the tasks in your own role and choose one verification, problem-framing or client skill that matters regardless of the model.
- Employment among US developers aged 22 to 25 fell nearly 20% since 2024. It is an early signal of pressure on entry roles, not proof that the profession is disappearing. Source: Stanford HAI, AI Index 2026, Economy. An early-career professional can show more than output: publish a short reproducible case with the task, sources, verification and what changed after use.
How to read it
These figures use different methods and time horizons. They do not add up to one forecast. Together they show something else: the cost of performing a task is falling faster than organizations are redesigning processes, roles and the way they measure results.
What it changes for people and work
- Professional: value moves from "I produce quickly" to "I understand the task, verify the work and own the result."
- Manager: lower production cost does not turn into profit on its own. Measure the process outcome, not the number of AI pilots.
- Founder: the opening is where the cost of production collapsed while customers still pay for a solved problem.
What it changes for teams and business
The outcome needs an explicit owner. AI can perform part of the work, but people choose the direction, verify the quality boundary and decide whether to act. "People own the outcome" is about agency and stewardship, not blame.
Working map for the week
| Decision | When it fits | First move | Boundary |
|---|---|---|---|
| Automate a step, not a whole role | The workflow is frequent and narrow | Map its agent, human and robot steps | Automate only a step with a quality criterion and outcome owner |
| Choose a skill that gains value | Work changes faster than job titles | Map your tasks and choose a verification, framing or client skill | Do not make a career forecast from one global figure |
| Check AI value | The team already uses a tool | Link output cost to an outcome measure | Do not call it savings when the result did not change |
Continue in the Practicum
Cheap output does not create an advantage on its own. Take one real task through the Volume 1 workbook: it separates cost savings that have become the norm from growth and trust that may remain distinctive. Then use the Volume 1 workbook before calling speed profit or redesigning a whole role.
One sensible next move
Pick one process where your team already uses AI. Write down the cost of producing the output and a separate measure of the real outcome: time to decision, conversion, error rate, revenue or risk removed. If the second measure is missing, you are tracking cheap output rather than valuable outcome.
Confidence and sources
Epoch AI, McKinsey, WEF and Stanford HAI are used for the factual signals with the limitations stated above. The output-to-outcome reading is the Practicum's editorial interpretation. Check the date: rates and estimates move.