The Manifest
The day's signal across AI models, supply chain, deals, research, and the org layer.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01Quoting Anthropic — 'As of May 2026, more than 80% of the code we merge into Anthropic's codebase was authored by Claude.' Mollick notes this matches…X
- 02A thread cataloging every AI agent failure of the past 60 days, anchored by the PocketOS incident — a Cursor agent running Claude Opus 4.6 deleted a…X
- 03Cites the MIT NANDA study finding only 5% of enterprise AI implementations deliver measurable returns.'Why AI Works Best When It Works with Humans'
- 04AWS bundled 25+ specialized supply chain tools into a set of agentic 'teammates' that orchestrate planning, scenario modeling, and exception triage.Amazon Connect Decisions (general availability)
- 05Schmitt, Vial & Blohm argue the winning GenAI org structure is a 'spine' — a cross-functional connective tissue between users, domain experts, and a…Summer 2026 issue
Models & Releases
5 storiesKarpathy at Sequoia AI Ascent 2026
fireside chat, week of Jun 2. Core thesis: LLMs aren't just speeding up existing work; 'Software 3.0' means prompts are the new programming language and 'neural computer' agent-native infra is coming.
Microsoft MAI model family
(Jun 2): MAI-Thinking-1, MAI-Code-1-Flash, MAI-Image-2.5 (above Gemini on Arena), MAI-Transcribe-1.5 (43 languages, 5x faster). Microsoft is openly de-coupling from OpenAI.
Gemini 3.5 Pro committed for June 2026 release
, with Flash already shipping at $1.50/$9.00 per M tokens and beating 3.1 Pro at 4x speed on agentic benchmarks.
Claude Opus 4.8
now resolves 88.6% of SWE-bench Verified vs GPT-5.5 at 88.7% — effectively tied at the coding frontier.
arXiv 2506.06576 'Future of Work with AI Agents'
36% of occupations have workers using AI for ≥25% of tasks (Anthropic usage data).
Supply Chain & Ops
4 storiesAmazon Connect Decisions GA
(carrying into June discourse). Agentic SC planning bundle.
McKinsey distribution-ops research
AI is delivering 20–30% inventory reduction, 5–20% logistics cost reduction, 5–15% procurement spend reduction at distributors.
BCG 'Supply Chain Planning 2026: Why AI Alone Isn't Enough'
argues that without process redesign, AI demand planning underperforms even classical methods. Direct counter-take to the agentic AI hype cycle.
SAP 'Autonomous Supply Chain' piece
(June): names the shift from copilots to 'coordinated agent-to-agent workflows.'
Org & AI Architecture
5 storiesHBR 'Why AI Works Best When It Works with Humans'
(Jun 1) — citing the MIT NANDA '5% of AI projects deliver measurable returns' finding.
MIT Sloan Summer 2026 'AI Spine'
framework — named organizational pattern for scaling GenAI.
Atlanta Fed / NBER w34984
'AI, Productivity, and the Workforce' — corporate-executive evidence: 11.5% net productivity gain, 4% net headcount decline over 12 months. Larger firms expect AI-driven workforce reductions; smaller firms expect modest gains.
arXiv 2604.03517 'The Augmentation Trap: AI Productivity and the Cost of Cognitive Offloading'
(Caosun & Aral) — empirical evidence that AI productivity gains carry a measurable cognitive-offloading tax. Aral's name = signal.
HBR 'AI Doesn't Reduce Work — It Intensifies It'
(Feb 2026, still circulating): employees compress more tasks, more hours, more cognitive load — productivity rises, burnout rises faster.