The Future of Software Will Be Shaped by Microeconomics with Tim O'Reilly

Why It Matters

Decisions about model placement, pricing and orchestration will determine who captures value in the AI stack and which business models scale, creating major opportunities for new platform vendors and changing how companies build software. Automating these economic trade-offs could dramatically lower costs and enable wider, more efficient deployment of AI capabilities.

Summary

Tim O’Reilly argues that the next phase of software will be driven by microeconomic choices about when to use conventional software versus large language models, and when to run models locally or in the cloud. He predicts a new discipline blending computer science with economic mechanism design to automate those trade-offs and optimize system-level costs and performance. O’Reilly compares this moment to past platform breakthroughs—like pay-per-click auctions and CGI—that unlocked massive commercial ecosystems, and says a wave of infrastructural and architectural innovation is imminent. He expects both startups and open-source projects to emerge that balance entire systems the way Databricks, Snowflake and Kafka reshaped data infrastructure.

Original Description

The software industry is being shaped by microeconomics. The discipline that's emerging focuses on choice and trade-offs: When should you use an LLM versus conventional software? When does a local or open weight model beat a frontier one? And is what the inference actually costs worth what it returns? As Tim O'Reilly points out, there's a parallel here to pay-per-click advertising, which started as a crude highest-bidder auction before Google refined it into something that actually worked. Or CGI, which let Rob McCool wire a web server to a database and quietly opened up everything that followed. We're at at a similar moment now. The models exist. The architectural innovation cycle that figures out how to use them well is just beginning. The companies that will matter, the way Databricks and Snowflake and Apache Kafka mattered, probably haven't been built yet.

Follow O'Reilly on:

LinkedIn: https://www.linkedin.com/company/oreilly/

Facebook: http://facebook.com/OReilly

Instagram: https://www.instagram.com/oreillymedia

BlueSky: https://bsky.app/profile/oreilly.bsky.social

TikTok: https://www.tiktok.com/@oreillymedia