Big-four Q1 2026 earnings split the market along the vertical-integration axis. Nadella publicly concedes Microsoft's eroding trust. The frontier-lab universe firms its boundaries. RAMageddon crosses into enterprise. The harness layer becomes a public governance object. Fifty-nine signals across five days; the post-Watershed economy hardens around its constraints.
Big-four Q1 2026 earnings disclosed an asymmetric tape along the vertical-integration axis; the analyst-class capex regime crystallises at $1.1T; and Nadella publicly concedes Microsoft's eroding trust on the same day.
Fourteen signals across three days. The substrate bifurcation is now priced in business days — DeepSeek V4 Pro landed on Ollama two days after release. The agent runtime is meeting hostile terrain at the database layer and on the open web. And the infrastructure capital is finding new routes through sovereign silicon, agentic CPU buildouts, and Apple's quiet RAM strategy.
Twenty-two signals across three days trace the same structural pattern: the partnerships, pricing models, and geopolitical assumptions that held the prior paradigm together are unbundling simultaneously — and the harness is what survives each unbundling.
Twelve signals from the long weekend converge on one structural fact: the substrate has become portable and the harness is all that matters. DeepSeek V4 ships with Anthropic API compatibility, Gas City specifies the team size to replace seven-figure SaaS bills, Google commits $40bn in cash and compute to Anthropic, Intel's 18A node yields, and xbow replicates Mythos-class offensive capability for everyone.
Mark Pesce names the ceiling of the good-enough zone. Mythos leaks through a contractor-adjacent breach and the NYT calls frontier launches weapons tests. OpenAI leads a launch on cost. Agent swarms become the default pattern. And usage and hostility rise together in the cohort entering the labour market. Sixty-six signals across five days; the post-Watershed economy is acquiring its boundaries in public.
Mark Pesce's fourth post-Watershed paper names the ceiling on harness safety; Mythos access leaks via a contractor-adjacent breach just as the NYT reframes frontier launches as weapons tests; OpenAI leads with cost; Anthropic publishes a postmortem; semis notch a sixteenth straight up-day; TenCent and Alibaba move on DeepSeek; and adoption and hostility toward AI are rising together.
Three independent announcements converge on multi-agent orchestration as the default pattern, while Google discloses that three-quarters of its code is now AI-generated — and a consulting class sells one-day AI credentials to enterprises that haven't started.
Salesforce declares 'the API is the UI', Figma's structural crisis deepens, and Amazon's $5B Anthropic investment reveals the circular logic of AI infrastructure — while Apple's new hardware-engineer CEO inherits a company that can't ship competitive AI software.
From Bloomberg's semiconductor doubling to Dario's open-source timeline, the infrastructure thesis hardens while the application layer above it fractures — and Chinese labs ship agent swarms that make the watershed personal.
The infrastructure layer keeps hardening — $581B in committed capital, GPU rental prices still climbing, Amazon locking up Australian energy — while the application layer above it compresses faster than anyone expected.
Tokens exist in three quality classes relative to any given harness: not good enough, good enough, and too good. The third class is new, and dangerous. Anthropic's Mythos Preview provides the first evidence that too-good tokens are no longer theoretical.
The harness hierarchy gets its formal paper. Open weights cross the watershed. Infrastructure declares itself at every layer from lithography to grid regulation. And the paradigm casualties — pull requests, Copilot, pre-watershed startups — keep accumulating.
The infrastructure stack declares itself at every layer — from ASML's 83% lithography monopoly to Australian grid standards for data centres — while the harness hierarchy gets its formal paper and open weights cross the watershed.
Seven signals today locate the seams that are now carrying the weight — runtime over training, harness over model, CPU over cloud, open disclosure or closed consortium — each seam attracting its own capital and its own attackers.
Seven signals today. Mark formalises the harness hierarchy as the demand-side equation of the post-watershed economy; cybersecurity becomes proof-of-work infrastructure; mathematics is the first formally verifiable domain to cross; agents are being plumbed across runtime, identity, payments, and language; and the analyst framework starts decomposing in public.
Seven signals today, one through-line: the post-watershed economy is being named out loud. Cyber is infra, agents are the primary API customer, the Lean Startup author is retiring his own framework, and the incumbents are being repriced in real time.
AI crosses genuine capability thresholds in cybersecurity as institutions — governance bodies, market analysts, and companies — remain anchored to pre-watershed frameworks. Meanwhile, the infrastructure anti-bubble case hardens, and the harness layer proves its primacy over raw model capability.
Fifty-eight signals in one week. The capability frontier outran its guardrails. The memory crisis closed a feedback loop. The open-weight tier crossed the watershed. And the agent runtime layer stopped being a research category and became a contested market.
Jevons paradox arrives for frontier models, AWS builds the agent registry, Intel's data centre revival at $1B custom silicon, and why AI adoption surveys are measuring the wrong tier.