The business AI market just had its first leadership change, and it happened faster than most expected. This edition we look at why Anthropic has overtaken OpenAI in enterprise adoption and what that tells us about how quickly incumbents can lose ground, how agentic AI is splitting data engineering into two fundamentally different jobs, and why the compute and power demands of running agents at scale are turning out to be a much bigger problem than anyone budgeted for.
The trend: Anthropic has overtaken OpenAI is business adoption for the first time, with 34.4% of businesses now using Anthropic versus 32.3% for OpenAI.
The details: According to Ramp's AI Index, Anthropic quadrupled its business adoption over the past year while OpenAI grew just 0.3%. This has been driven largely by Claude Code's dominance in coding workflows. But Anthropic's token-based pricing model gives it an incentive to push businesses toward more expensive models even when cheaper ones would do, with Uber's CTO recently revealing the company already blew through its entire 2026 AI budget just on Claude.
Why it matters: The company leading business adoption today can lose that lead within months, and with switching costs low and model quality converging, enterprise teams should think carefully before building around any single provider.
The trend: 62% of organizations using AI-driven orchestration have already seen a 40% or greater reduction in pipeline maintenance time.
The details:Writing for HackerNoon, a data engineering lead at AWS outlines that there are now 3 categories of work within data engineering. Tasks agents now own entirely, tasks where humans and agents collaborate, with the human making the final call, and tasks that are growing in importance precisely because agents are being deployed, like designing governance frameworks and evaluating whether agent behavior can actually be trusted. As a result, data engineering is splitting into two distinct roles: the systems engineer who owns the infrastructure agents run on, and what the author calls the intelligence engineer, focused on designing architectures, building frameworks, and governing agent behavior.
Why it matters: Most teams today are still hiring for roles that agents are actively absorbing. Are data teams restructuring around these changes or just adding AI tools to workflows designed for a bygone era? In an ideal world, agents should free people up to work on activities that will drive growth, like marketing spend optimization or inventory management models.
OpenAI just launched a personal finance feature inside ChatGPT, letting US based Pro users connect their bank accounts, investments, and credit cards to get spending analysis, budgeting plans, and planning grounded in their actual financial data rather than generic advice. The shift is a big one: without connected accounts you get a list of generic saving tips, with them you get a breakdown of exactly where your money went last month and a specific monthly plan. The obvious concern is handing an AI model a complete picture of your financial life, and whether the productivity gains are worth the privacy tradeoff. But for people already piecing together their finances across five different apps and a spreadsheet, having one place that actually knows the full picture may be worth it.
Stanford's Digital Economy Lab found that agentic coding tasks consume 1,000 times more tokens than standard chat queries, with token usage varying by up to 30 times across runs of the same task. AMD has revised its server CPU market growth forecast from 18% to over 35% annually as a result, with Morgan Stanley projecting that agentic AI adds between $32.5 billion and $60 billion in new CPU demand by 2030. Goldman Sachs puts the broader picture in stark terms: token consumption is projected to grow 24 times by 2030 compared to 2026 levels, and data center power consumption is expected to jump 175% over the same period, with agents as a primary driver.
%201.svg)
