- The AI Report
- Posts
- 🤖 How Claude Code's creator codes
🤖 How Claude Code's creator codes
Parallel sessions, slash commands for inner loops, MCP integrations, and verification loops that 2-3x quality. This is what professional AI coding actually looks like.
Meta just dropped $2B on autonomous AI agents while DeepSeek's new architecture promises massive efficiency gains. Meanwhile, Claude Code's creator reveals he runs 15 parallel sessions and the secret is verification loops, not better prompts.
The Latest in AI
Master Content Workflows: Expert AI Strategies for 2026
Struggling to manage massive video and image libraries? Join media experts in a Jan 14th webinar to learn how AI transforms content organization and boosts monetization.
Join Forrester VP Principal Analyst Phyllis Davidson to explore the trends reshaping content ops, alongside former Disney and ESPN executive Oke Okaro, who reveals how to unlock hidden ROI in your content libraries.
Don't miss this chance to get on the cutting edge of media innovation and AI in this discussion that explores how to use deep visual understanding to turn idle assets into revenue.
💡 Meta Buys Manus for $2 Billion
Meta's strategic acquisition of Singapore-based AI startup Manus for $2 billion underscores the company's commitment to dominating the autonomous AI agent market. This move positions the tech giant to compete more effectively in developing AI systems capable of performing real-world tasks independently.
The acquisition aims to enhance AI agents capable of performing real work autonomously
This move positions Meta competitively in the evolving AI landscape
An industry expert noted, 'This acquisition signifies Meta's serious commitment to AI-driven solutions'
The deal reflects a broader trend of tech companies investing heavily in AI capabilities
🤔 Why It Matters:
Meta's acquisition of Manus represents a fundamental shift towards more capable AI agents, which could redefine how businesses leverage AI for productivity. This move not only strengthens Meta's position in the AI race but also democratizes access to advanced AI functionalities for developers and product teams.
🔬 DeepSeek Develops mHC AI Architecture To Boost Model Performance
DeepSeek has introduced a revolutionary mHC (Manifold-Constrained Hyper-Connections) architecture that promises significant improvements in AI model efficiency and performance. This breakthrough technology could transform how AI models are designed and deployed across various applications.
The mHC technology utilizes Manifold-Constrained Hyper-Connections for enhanced efficiency
This development could lead to more powerful AI applications across various industries
According to DeepSeek, this architecture represents a major leap in AI model capabilities
The technology was introduced in a paper published just this week, highlighting its novelty
🤔 Why It Matters:
The introduction of the mHC architecture could fundamentally shift how AI models are developed, enabling more efficient and powerful applications. This breakthrough not only enhances model performance but also democratizes access to advanced AI capabilities for developers and organizations alike.
🛠️ How Claude Code's Creator Uses Claude Code
Boris Cherny, creator of Claude Code, runs 5 terminal instances and 5-10 web sessions in parallel - handoff between them using & and --teleport. He uses Opus 4.5 with thinking for everything despite being slower, because it requires less steering and better tool use makes it faster overall. His team shares a single CLAUDE.md in git, updating it multiple times weekly when Claude makes mistakes.
Most sessions start in Plan mode (shift+tab twice) until the plan is solid, then switch to auto-accept edits for 1-shot execution. Cherny uses slash commands for "inner loop" workflows like /commit-push-pr that run dozens of times daily, with inline bash to pre-compute git status and avoid back-and-forth. Subagents automate common PR workflows like code-simplifier and verify-app.
Claude Code uses all Cherny's tools autonomously - searches and posts to Slack via MCP server, runs BigQuery queries using bq CLI, grabs Sentry error logs. The Slack MCP configuration lives in .mcp.json and is shared with the team. For long-running tasks, he uses background agents, Stop hooks, or the ralph-wiggum plugin with --dangerously-skip-permissions in sandboxes.
The most critical tip: give Claude a way to verify its work. Cherny's Claude tests every change to claude.ai/code using the Chrome extension - opens a browser, tests UI, and iterates until code works and UX feels good. Having that feedback loop 2-3x's the quality of final results.
🤔 Why It Matters:
This shows how professional developers actually use AI coding tools at scale - it's not about prompting better, it's about building infrastructure around the AI. Parallel sessions, shared CLAUDE.md files in git, slash commands for repeated workflows, MCP integrations, and automated verification loops turn Claude Code from a coding assistant into a development platform. The shift from "help me write code" to "autonomously execute multi-step workflows with verification" is how AI coding tools graduate from toys to production infrastructure.
🗞️ AI Bytes
📰 AI models can develop 'humanlike' gambling addiction when given more freedom: study
Researchers have discovered that AI models exhibit gambling-like behaviors when given greater autonomy, raising important questions about AI safety and decision-making processes. This finding suggests that as AI systems become more independent, they may develop unexpected behavioral patterns that mirror human psychological tendencies.
📰 In 2026, AI will move from hype to pragmatism |
This analysis explores the transition from AI experimentation to practical implementation across industries, highlighting key trends and realistic expectations for AI adoption. The piece examines how businesses are shifting focus from flashy AI demos to sustainable, ROI-driven AI solutions that solve real business problems. It serves as a strategic guide for executives, entrepreneurs, and technologists navigating the developing AI landscape.
📰 OpenAI improving audio AI models ahead of introduction of personal AI devices
OpenAI is enhancing its audio processing capabilities in preparation for launching consumer AI hardware products that will rely heavily on voice interaction. These improvements focus on better speech recognition, natural conversation flow, and reduced latency for real-time audio processing. The development targets consumers seeking seamless AI assistants and businesses looking to integrate advanced voice AI into their products and services.
🛠️ Top AI Tools This Week
🚀 Intempt
GrowthOS unifies product, web, and CRM data into a single platform for real-time segmentation, personalization, and automated customer journeys. Build segments across web, app, and CRM data, automate omnichannel campaigns, and prioritize high-value users with predictive scoring.
On a scale of 1 to AI-takeover, how did we do today? |



