How the engineer behind Claude Cowork actually uses Claude

#1 𝕏

Qwen launched implicit caching on Qwen3.7-Max—automatic, zero-setup boosts in speed and cost efficiency. For higher, more deterministic hit rates, it recommends using explicit caching with published best practices.

#2 𝕏

xAI launched Grok Build Beta for SuperGrok and X Premium+ users, offering Plan Mode, Imagine for image and video creation, and a CLI to build automations and orchestrators.

#3 ▶️

How the engineer behind Claude Cowork actually uses Claude | Felix Rieseberg (Anthropic)

How I AI Podcast

#4 📝 PromptLayer Blog

MCP vs API: Architecture Patterns for AI Agents and Applications - Discusses the protocols powering AI workflows—MCPs and APIs—explaining how both are used behind agent actions, data lookups, and prompt evaluations in modern AI systems.

#5 𝕏

Santiago demonstrates replacing token-burning Claude Code and Codex automations with simple scripts via Zapier’s SDK and explains when to use the SDK versus Zapier’s MCP.

#6 in

Greg Isenberg lays out a playbook for a cash‐flowing vertical AI agent startup: manually map a boring industry workflow—interview 10 daily users, document every step and edge case in Obsidian—then automate it using Hermes, Obsidian Vault, Composio, Claude Code/Codex, Perplexi...

#7 in

Udi Menkes dramatically boosted AI performance by writing a 50-line markdown “resolver” that maps each task to the three most relevant “brain” files, solving context overload overnight.

#8 📝 Simon Willison

Microsoft Copilot Cowork Exfiltrates Files - A report describes how Microsoft Copilot Cowork allowed agent-sent emails to leak data via externally rendered images and pre-authenticated OneDrive links, creating a path for prompt-injection exfiltration. The post highlights the continued challenge of designing agentic systems that don't enable attackers to extract sensitive data.

#9 𝕏

LlamaIndex 🦙 added native HEIC support to LlamaParse, so you can point it at Apple’s default image format—whiteboard pics, scanned docs, receipts—without converting to JPEG first.

#10 📝 PromptLayer Blog

Braintrust Alternatives — The Best Prompt Management Platforms in 2026 - A buyer-focused piece for teams evaluating Braintrust that covers operational considerations like tracing volume, evaluation cost, and speed of shipping changes when choosing a prompt management platform.

#11 📝 Ampcode Chronicle

GPT Image 2 Paints Better - Amp's painter tool now uses GPT Image 2, which Amp claims outperforms Gemini 3 Pro Image at preserving existing text, typography, and visual style when editing UI screenshots while costing roughly one-quarter the price. In an example thread, Painter converted a screenshot of the Chronicle page into an updated design while retaining its original visual style.

#12 ▶️

Codex 5.5 vs Claude Opus 4.7 Polymarket Trading Challenge

All About AI

#13 𝕏

There's An AI For That reports that @Speechmatics tops voice AI benchmarks—offering 25% higher accuracy than most competitors plus sub-second latency in 55+ languages—underscoring that fast but wrong is still wrong.

#14 𝕏

Peter Yang shares Ryan Carson’s top lesson: investing in documentation, cron-driven skill files and foundational systems for AI agents—instead of just slapping together a bare-bones MVP—unlocks the equivalent output of 10 people and enables shipping 10 PRs a day.

#15 𝕏

Lenny Rachitsky: Dan Shipper now does all his writing, research and email inside AI agents like Codex or Claude Code—using Google Docs, PostHog and other tools in the agent’s in-app browser for seamless, context-rich collaboration.

#16 𝕏

clem 🤗 ran thousands of queries and used @DAKlingbeil’s Submarine.ai to analyze how coding assistants mention Hugging Face products (see JSONL dataset). They’re asking the community for alternative or more effective analysis approaches.

#17 𝕏

clem 🤗 warns the most important AI risk is the concentration of power, capabilities, and economic gains.

#18 in

Anu Jagga Narang argues that AI can speed up creating strategy artifacts but can’t replace the critical thinking of deciding what not to pursue. Her new High Agency PM post uncovers the strategy section most teams leave blank and shows how AI exposes that gap.

#19 in

John Cutler warns that companies without solid causal models for their investments before AI won’t suddenly figure one out now, and will just waste time measuring the wrong things.

#20 in

Arvind KC warns that AI models are simplifying expert tasks but producing a flood of homogeneous output. He argues true differentiation will come from human framing, judgement, taste, and systems-building.

#21 𝕏

claire vo 🖤 After 20 years of coding, Codex took over: planning and executing a full core-app refactor (publishing HTML docs, running iterative loops with browser smoke tests, linting and regression fixes) and even cleaning out 4,000 emails, leaving me to just ask “OK, what’s...

#22 𝕏

Logan Kilpatrick announced that Lyria 3 is now available via the API, enabling developers to build with it.

#23 𝕏

Santiago Oracle’s DevRel team pushed extensive free tutorials and resources to FreeSQL.com and published a host of AI-integration examples in their oracle-ai-developer-hub GitHub repo.

#24 𝕏

Garry Tan reports that Alibaba’s Qwen2.5-7B Instruct matches GPT-3.5-turbo performance, showcasing the rising quality of local open-source LLMs.

#25 𝕏

Marily Nika tested her AI agent “Claw” over United’s text-only in-flight messaging, proving AI agents can run smoothly even with minimal bandwidth.