Video summary

Pricing and Productivity: The Economics of AI

Main summary

Key takeaways

News and Commentary

Summary: “Pricing and Productivity: The Economics of AI”

The event is an emerging tech economics discussion hosted by the Federal Reserve Bank of San Francisco. Speakers explore how AI affects productivity, pricing, and labor-market dynamics, using Amazon Web Services (AWS) as a lens for how AI-enabled business infrastructure is built and deployed.

1) Why AI infrastructure is booming—and what AWS is doing about it

The discussion highlights that data center investment (and broader spending on information-processing equipment) is rising rapidly as firms pursue AI capability. The pace is described as stronger than prior tech-boom periods (e.g., the mid-1990s to mid-2000s).

  • Energy constraints and large-scale capital needs are likened to:
    • electricity requirements, and
    • airline/hotel-style amortization (big upfront capacity, then efficient allocation over time).

AWS’s response is described as twofold:

  1. Investing in infrastructure to meet accelerating demand.
  2. Developing AI services on top of cloud compute and storage—helping customers deploy AI faster and more effectively.

Examples of AWS tools/products mentioned include:

  • “Q” (coding assistant) and QCLI
  • Quick Suite (collaboration/workflow-oriented tools)
  • QuickSight (turning internal documents/knowledge into a queryable “knowledge base”)

2) AI is reducing software-development “fixed costs,” potentially accelerating productivity

A central argument is that genAI can:

  • compress development timelines, and
  • reduce costs of building and iterating software.

An internal example is given:

  • Work that previously took ~3 months with non-genAI approaches was reduced to ~2 weeks using genAI tooling.

The speakers connect this to macroeconomic themes:

  • Historically, adopting new technology involves learning/implementation fixed costs.
  • AI—especially alongside cloud—may make adoption and experimentation faster.
  • The cloud transition is also framed as shifting from CAPEX to OPEX, lowering barriers to capacity and experimentation.

3) Productivity measurement is hard; evidence is mostly “partial equilibrium”

When asked about research evidence (including randomized controlled trials), Jacob Lavrijsen says AWS internal findings broadly align with academic work, including:

  • AI tends to increase “velocity” (e.g., faster output on software development tasks), using measurable metrics.

He also explains why economy-wide productivity is difficult to estimate:

  • Productivity effects across industries require general equilibrium reasoning, which is harder to quantify.
  • Therefore, many results are best interpreted as lower-bound or partial-equilibrium findings.

Overall: measurable improvements may not straightforwardly translate into confirmed aggregate productivity gains.

4) Guardrails and “interrogability” to ensure outputs are correct

A recurring theme is that AI’s value depends on being able to trust and audit results.

The discussion describes an “interrogable” component as one where:

  • outputs include citations that can be checked,
  • results can be verified by examining underlying queries/executions (e.g., agent-generated SQL run against databases),
  • the analysis can be validated rather than treated as a black box.

5) Entry-level jobs and skill requirements: hiring and adoption patterns

A question asks whether AI is reducing opportunities for entry-level workers. The response suggests:

  • At the economy level, patterns like graduate unemployment may reflect multiple concurrent factors, not only AI.
  • Within AWS, Lavrijsen says his organization is hiring more younger people, attributed to:
    • post-COVID hiring shifts, and
    • organizational “shape” (avoiding being overly top-heavy).

He argues that younger workers may be “native” to AI/code assistants, potentially boosting productivity and speed, while experienced staff remain important for guidance.

6) Pricing: AWS perspective focuses on customer understanding, not real-time dynamic pricing

Asked about AI and dynamic pricing, Lavrijsen says:

  • his team is not working on real-time generative-AI pricing.

Instead, they use AI to:

  • understand customer preferences, and
  • experiment to identify what creates measurable customer value (“what good looks like” for customers).

7) Final takeaways: rapid pace, fast internal diffusion, early days

Lavrijsen highlights several “surprise” points:

  • The speed of development of AI tools and services.
  • The within-company diffusion of AI after adoption—making downstream measurement harder due to:
    • timing effects, and
    • measurement error.
  • The ability to change workplace modalities, such as:
    • querying knowledge bases, and
    • enabling faster collaboration.

He characterizes the stage of adoption as still very early for both cloud services and genAI.

Presenters / Contributors

  • Louise Willard (Executive Vice President and Chief Information Officer, Federal Reserve Bank of San Francisco)
  • Sylvain Leduc (Executive Vice President and Director of Economic Research, Federal Reserve Bank of San Francisco)
  • Jacob Lavrijsen (Chief Economist, Amazon Web Services)

Original video