Tech • IA • Crypto
Anthropic Claude Co-work, Devin, and AI agents reshape coding
Briefing
Claude Co-work transforms legal ops
Anthropic is positioning Claude Co-work as a legal “chief of staff” that automates daily triage and briefing. It compiles updates from Gmail, Slack, and internal systems into prioritized summaries, reducing manual review. The /brief command retrieves targeted context from long documents, often exceeding 40 pages, surfacing only relevant excerpts. Human validation remains central before decisions are made, preserving accountability.
AI agents accelerate enterprise engineering
Autonomous agents are moving from assistance to execution across real systems. Spotify reports agents generating over 1,000 pull requests per month, cutting migration timelines by more than 90%. In public services, Binti reduced foster care licensing by 20 days using AI workflows. These deployments show measurable operational gains rather than experimental pilots.
Devin pushes autonomous coding frontier
Devin exemplifies a shift toward fully autonomous AI software engineers capable of planning and executing tasks end-to-end. The goal across the sector is up to 10× faster development cycles. Adoption spans banks, health insurers, governments, and private equity firms, signaling broad enterprise demand. The competitive landscape suggests a multi-tool ecosystem rather than a single dominant platform.
Context window limits shape coding
AI coding systems are constrained by finite context windows that store prompts, tool calls, and outputs. As these fill, performance and accuracy can degrade without intervention. Developers must actively manage this resource to maintain reliability in complex workflows. The constraint is emerging as a core design consideration for agentic systems.
Compaction trades memory for capacity
Automatic and manual compaction compress prior interactions to free space while preserving key information. Commands like /compact summarize history and discard low-value details such as verbose tool logs. This introduces a trade-off between fidelity and continued progress. Poor compaction can remove subtle but important context needed later.
Reset and inspect context usage
Developers can fully reset sessions using /clear, eliminating residual bias from earlier tasks. Tools to inspect usage, such as /context, provide visibility into how much capacity remains. These controls are becoming standard in advanced coding environments. They enable deliberate workflow boundaries between features or projects.
AI standardizes marketing analytics workflows
AI systems are automating weekly reporting by aggregating and validating data across sources. Structured “skills” encode repeatable steps like report generation and proofreading, ensuring consistency. Inputs from Slack, meeting transcripts, and data warehouses are synthesized into ready-to-review drafts. Human teams still decide narrative focus and strategic direction.
Sales teams gain unified account intelligence
Claude Co-work is also applied to sales, unifying fragmented data from Salesforce, email, Slack, and warehouses. Custom “skills” define what signals to gather and how to present insights for each account. The system produces pre-meeting briefs and automates follow-ups, replacing hours of manual research. This integration enables faster, more informed customer engagement.
Videos covered
Context Management in Claude Code
- •Finite context window as a core constraint
- •Automatic and manual compaction
- •Risk of information loss during compaction
Claude Cowork for legal teams
- •AI as a Legal “Chief of Staff”
- •Rapid Context Retrieval with /brief
- •Integration Across Workflows
Claude Cowork for marketing ops
- •AI streamlines marketing analytics workflows
- •Structured “skills” standardize recurring processes
- •Automated weekly preparation across systems
Claude Cowork for sales
- •Fragmented data challenge
- •AI-generated account intelligence
- •Integrated data access
The Problem Solvers: Scott Wu at Cognition
- •Rise of AI Software Engineers
- •Mission to Accelerate Development
- •Foundations in Competitive Programming
Code with Claude London 2026: Opening Keynote
- •From hobbyist coding to autonomous systems
- •AI agents delivering real-world impact
- •Rapid acceleration in model capabilities