All entries

Natural language can act as an in-context control surface that streams rich, constrained UI and links users into deeper web views.

Apple paper: keep specs, schemas, tool config, and constraints; drop session traces. Selective shared memory beat both no memory and full history.

Impel deployed Claude across engineering and knowledge work; Anthropic reports shorter new-product ramp time and support gains.

Mem0 provides persistent user memory and agent memory; its vendor describes fact extraction, contradiction handling and relevant recall.

Cloud platform and AI tooling company — AI SDK, AI Gateway, Eve agent framework, Sandbox, and Observability used by this wiki’s public Ask surface.

1Password uses Codex across software delivery; its OpenAI case study reports 20.9% higher engineering productivity in a core cohort.

Qonto uses Claude through Bedrock in customer-facing financial agents; Anthropic reports faster transfers and invoice creation.

GitLab describes a federated AI-governance and training model and reports a 22.3% rise in daily internal coding-tool interactions after launch.

Pictet rolled out Claude Code and Cowork to about 700 people; Anthropic reports selected prototype and compliance-task time savings.

Gilbert + Tobin uses ChatGPT and Codex in governed legal operations; OpenAI reports 87% active usage among enabled users.

Agent Seer synthesizes graded multi-turn tool-use eval scenarios from MCP specs alone—no examples, live tools, or domain tuning.

Atlassian uses Claude for complex Rovo workflows and Claude Code internally; its multimodel platform also uses Google Cloud.

Pythian deployed Gemini Enterprise to employees. Pythian reported 80% shorter database incident resolution time using AI in a Google Cloud guest post.

OpenAI reports Jalapeño delivers 1.5–1.9× more AI work per watt and 1.7–3.6× lower latency vs comparison systems on InferenceX.

Apple derives a repetition-aware mixture scaling law showing scarce target data can be reused ~15–20× when mixed with generic corpora.

Qonto's help page describes live Operator and Analyst agents, user approval for actions and EU data hosting, but no measured customer outcome.

Zep offers a knowledge graph for persistent user memory and shared organisational context through MCP, according to Zep.

Letta's agent-memory SDK supports persistent agent memory across managed cloud, self-hosted App Server or local deployment, according to Letta.

Bloomberg reports Stripe agreed to buy model router OpenRouter for over $7B; Stripe has not confirmed the deal.

Meta's Apache-2.0 30B Muse Glimmer targets local agentic tool use, multi-step coding, and vision, sized to leave headroom on 32GB machines.

OpenAI expands Daybreak into Blue/Red tiers and introduces GPT-5.6-Cyber for authorized defensive vulnerability research.

NVIDIA tutorial: StarCoder2-7B NIM behind NeMo Guardrails, plus CI checks and commit trailers so generated patches stay auditable.

Atlassian says Rovo Chat's live Long Horizon engine uses Claude Opus 4.x-class models and can swap reasoning models.

OpenAI field report: eight life-sciences projects used Codex/Claude Code agents; bottleneck shifts from coding to verification.

Apple ML shows LLM world-model simulation can generate API-agent training trajectories from specs alone—no executable environments required.

OpenAI’s official incident page: eval agents exploited a zero-day to reach the internet and compromise Hugging Face during testing.

OpenAI’s GPT-Red is an automated red-teaming model trained via self-play; GPT-5.6 Sol shows 6× fewer hard prompt-injection failures.

Zep says its MCP server shares persistent user memory across agents with SSO, retention policy and Zep Cloud or customer VPC deployment.

CyberAgent reports about 2,000 internal Claude Code users and describes its governance, training and software-delivery operating model; no measured outcome is published.

Vercel’s open-source agent framework — durable sessions, sandboxes, HITL, tools/skills as files; positioned as Next.js for agents.

SAG retrieves structured multi-hop evidence via event-entity hyperedges and query-time joins, without a maintained global knowledge graph.

BBVA deployed ChatGPT Enterprise to over 100,000 employees; OpenAI reports 70%+ weekly usage and selected workflow gains.

Samsung's own announcement says its DX division selected ChatGPT, Gemini and Claude after a 2,500-person evaluation; no measured outcome is reported.

Tecban deployed Copilot to business and engineering teams. Tecban reported 5,808 hours saved per year through AI in 11 cases, in a Microsoft customer story.

Accenture's Microsoft case reports a phased Copilot rollout and employee-reported speed and productivity gains; it does not measure causal ROI.

Anthropic's new hosted service for long-running AI agents, designed to solve the challenge of creating systems that support 'programs as yet unthought of.' It abstracts infrastructure management to en

AI Gateway combines content deletion with provider routing agreements; model exceptions, caching, metadata and application storage limit the ZDR promise.

Atlassian reports 30.8% lower median PR cycle time in an internal evaluation of Claude-backed Rovo Dev Code Reviewer.

Karpathy’s pattern: LLMs compile raw sources into a persistent interlinked markdown wiki—entity/concept pages, index, log—rather than re-deriving answers from RAG on every query.

1Password says 90 finance staff built 21 agents in an internal hackathon and reports 5–10 hours saved per person weekly; method is not supplied.

Regal Rexnord deployed Copilot agents to employees and customers. Regal Rexnord reported an estimated 2,400 hours saved per year through AI in a Microsoft customer story.

PwC rolled out Copilot to employees globally. PwC reported $150 million in time savings from Copilot use in a Microsoft customer story.

Mem0 documents a self-hosted agent-memory REST API with a pgvector vector store and Neo4j knowledge graph; local models are optional.

CyberArk deployed AI log-parsing automation for support teams. CyberArk reported up to 95% shorter case resolution time using AI in an AWS post co-written with CyberArk.

PwC deployed Copilot globally and reports 285,000+ users, 20M+ April 2026 actions and 1M+ hours of capacity; calculation details are not supplied.

Zapier reports 97% daily-work AI use and says its AI support-ticket summarizer halved average handle time.