Anthropic's terminal coding agent: CLAUDE.md, hooks, skills, subagents, plan mode.
As models improve, their baseline capability handles simpler tasks without needing explicit logs of past failure modes. Furthermore, maintaining these context logs across different model versions can end up overconstraining the agent.
▶ 30:26
Claude Code’s Bitter Lesson: Prompts, Harnesses, Mods, and Agents — Thariq Shihipar, Anthropic
The terminal interface started completely by accident because it was the cheapest and easiest way to get a prototype up and running without building a UI. The creator just wanted a simple way to test the API and bash tool use, but it quickly gained traction and stayed there.
▶ 4:23
Inside Claude Code With Its Creator Boris Cherny
You can use the advisor strategy to separate execution from advising. A smaller, cheaper model handles execution while a larger model like Opus acts as an advisor when needed. This approach delivers frontier-level quality at a fraction of the cost.
▶ 22:22
Code with Claude London 2026: Opening Keynote
Coding agents evolved from traditional chatbots by incorporating tool calling capabilities that let models execute shell commands and file edits. Instead of a strict human-model conversational turn, the model autonomously chains multiple tool calls to accomplish complex objectives. This transforms a simple prompt-response system into an autonomous execution loop.
▶ 4:01
Anthropic's Secret to Better AI Coding Agents
Claude Tag provides an extra degree of proactivity by autonomously initiating conversations, managing cross-channel memory, and executing long-running tasks. It can spin up its own sandboxes and ping you when a pull request is ready, shifting workflows from single-player sessions to asynchronous, multiplayer collaboration.
▶ 4:02
Inside Anthropic: How Claude Tag Is Changing Agentic Work
You should provide your agent with an accurate environment including emulators, profilers, and access to backend services. The Android CLI helps interface with various Android tools, manage SDKs, and connect to crash reporting.
▶ 4:01
Agentic engineering with Claude Code and Android CLI
Artifacts feature an associated database that stores persistent project data which multiple agents can access and update. They serve as an on-the-fly interface into the harness, helping surface rich details and requirements from the user.
▶ 8:52
Claude Code’s Bitter Lesson: Prompts, Harnesses, Mods, and Agents — Thariq Shihipar, Anthropic
Keep your instructions minimal and focused only on the absolute essentials like team PR workflows. If it expands to thousands of tokens and starts causing issues, you should simply delete it and start fresh. As models improve, you need fewer instructions to keep the model on track.
▶ 10:05
Inside Claude Code With Its Creator Boris Cherny
You can use MCP tunnels to connect Claude Managed Agents to internal servers behind your private network firewall. This routes requests securely without exposing your MCP servers over the public internet. This setup can be configured directly in the cloud developer console.
▶ 24:47
Code with Claude London 2026: Opening Keynote
The core edit tool relies on a primitive find-and-replace mechanism requiring exact byte-for-byte string matches without a cursor or selection interface. It takes parameters for the file name, old string, and new string directly from the model. If multiple identical old strings exist, the model must explicitly declare its expectations or the tool execution fails.
▶ 7:42
Anthropic's Secret to Better AI Coding Agents
Developers move away from single-player IDE sessions toward asynchronous, multiplayer collaboration directly inside communication channels like Slack. Instead of stepping through every turn of the agent, developers define success criteria and let long-running agents execute complex tasks and ping teams for reviews.
▶ 11:33
Inside Anthropic: How Claude Tag Is Changing Agentic Work
Keep your conversations around 150,000 tokens to get the full power of the model without context rot. You should also watch out for verbose logs like Gradle output by running them in quiet mode or letting harnesses truncate them.
▶ 9:12
Agentic engineering with Claude Code and Android CLI
Effective prompting requires building an accurate mental model of how Claude thinks, including what it can and cannot one-shot. Users must also learn domain vocabulary to reduce ambiguity and tackle unknown unknowns.
▶ 18:03
Claude Code’s Bitter Lesson: Prompts, Harnesses, Mods, and Agents — Thariq Shihipar, Anthropic
Multi-agent swarms allow multiple agents to operate with fresh, uncorrelated context windows, acting as a form of test-time compute. An engineer can set up a spec and an Asana board, allowing independent agents to pick up tasks and build features over a weekend without intervention.
▶ 22:07
Inside Claude Code With Its Creator Boris Cherny
Yes, you can configure Claude Managed Agents to use self-hosted sandboxes instead of public infrastructure. When an agent needs to execute work, it queues a work item that is picked up by your chosen provider such as Vercel or Modal. This allows you to run tasks securely on your own servers.
▶ 24:35
Code with Claude London 2026: Opening Keynote
Professional software engineering relies heavily on institutional knowledge, team chat discussions, CI results, and production dashboards rather than source code alone. Constraining an agent to a single repository folder creates severe performance limitations because it lacks access to real enterprise context. Giving models access to the broader work environment is essential for handling large-scale software engineering tasks.
▶ 15:04
Anthropic's Secret to Better AI Coding Agents
Agent execution time horizons are on an exponential trajectory, doubling roughly every four months according to METR research. This increase in capability means agents can handle more multi-step tasks, verify their own work, and earn greater human trust.
▶ 15:55
Inside Anthropic: How Claude Tag Is Changing Agentic Work
Use structured workflows starting with a skill that grills you for requirements to eliminate ambiguity. This breaks down a project into smaller issues that fit neatly within the context window limit before you begin building and verifying.
▶ 10:57
Agentic engineering with Claude Code and Android CLI
Task complexity dictates how effort scales, yielding major performance gains for security tasks where thorough edge-case testing is vital. Conversely, simpler software engineering tasks can use lower effort settings to conserve tokens since models require less verification overhead.
▶ 25:18
Claude Code’s Bitter Lesson: Prompts, Harnesses, Mods, and Agents — Thariq Shihipar, Anthropic
Plan mode is a prompt instruction with a limited lifespan that prevents the model from writing code prematurely. It works by simply adding a sentence to the prompt telling Claude not to code yet, letting it map out tasks before execution.
▶ 25:08
Inside Claude Code With Its Creator Boris Cherny
Claude Code provides multiple interfaces including the terminal CLI, IDE extensions, the full-screen desktop app, and the new agent view in the CLI. These surfaces allow you to track running, blocked, or completed agents across local and cloud sessions in one place.
▶ 33:51
Code with Claude London 2026: Opening Keynote
You can significantly improve agent performance by implementing tighter feedback loops using post tool use hooks that inject real-time errors directly into the tool response. By feeding information like type checking errors and linter warnings back immediately, the model receives a nudge at the exact moment of the mistake. This localized feedback loop beats relying solely on a smarter model or waiting for later compilation.
▶ 23:30
Anthropic's Secret to Better AI Coding Agents
Claude Tag agents are given their own independent permissions and API keys rather than assuming the user's permissions. This allows them to work autonomously on behalf of the team instead of an individual. Consequently, it becomes much easier to audit the agent's actions securely.
▶ 34:15
Inside Anthropic: How Claude Tag Is Changing Agentic Work
Keep instructions short and focused on things that always apply, such as architecture rules, module boundaries, and common code review mistakes. Avoid letting agents modify their own instruction files since it degrades performance and increases costs.
▶ 12:34
Agentic engineering with Claude Code and Android CLI
Maintaining different configuration files across various model versions is painful and can overconstrain Claude as models improve. In the limit, smarter models handle simpler tasks better natively, making these persistent markdown files largely unnecessary for new projects.
▶ 30:15
Claude Code’s Bitter Lesson: Prompts, Harnesses, Mods, and Agents — Thariq Shihipar, Anthropic
You should focus on serving the latent demand of the model by observing how it naturally wants to interact with the world and use tools. Rather than boxing the model into rigid APIs, you should enable its desired workflows and build for the model capabilities expected six months in the future.
▶ 31:22
Inside Claude Code With Its Creator Boris Cherny
Autofix listens for events like code review comments, security reviews, and CI failures, then proactively addresses them. It can rebase on merge conflicts or retry timed-out jobs so your pull requests stay green without manual babysitting.
▶ 36:22
Code with Claude London 2026: Opening Keynote
Modern LLM token prediction requires previous tokens to match, making dynamic cache eviction like traditional LRU caches extremely expensive. Swapping rules in and out costs significantly more because prediction costs multiply with cache misses.
▶ 30:48
Anthropic's Secret to Better AI Coding Agents
Anthropic found that opinionated memory abstractions and indexed stores were often too restrictive and prone to issues. As models became more capable, agents performed better using native bash, grep, and straightforward file systems. Removing complex memory structures simplified harnesses while letting models handle data effectively.
▶ 36:56
Inside Anthropic: How Claude Tag Is Changing Agentic Work
Use hooks to subscribe to agent events like session start, session stop, or file reading and writing. Instead of relying on prompt instructions to run linters or protect secret files, deterministic scripts executed via hooks reliably enforce these constraints.
▶ 15:35
Agentic engineering with Claude Code and Android CLI
Claude mods run in-process inside the TypeScript runtime rather than merely triggering external scripts on events. This allows them to access runtime scope details like turn counts and token usage, spawn sub-agents, and dynamically modify the UI.
▶ 42:03
Claude Code’s Bitter Lesson: Prompts, Harnesses, Mods, and Agents — Thariq Shihipar, Anthropic
Productivity measured by pull requests has grown significantly, with output per engineer increasing substantially since Claude Code was launched. Many engineers and creators now rely on it completely, uninstalling their traditional IDEs and writing 100% of their code through Claude Code.
▶ 41:00
Inside Claude Code With Its Creator Boris Cherny
Routines are higher-order prompts that can be configured to run on a schedule, via webhooks, or through API requests. They enable developers to delegate async tasks like resolving overnight GitHub issues and waking up to ready-to-merge pull requests.
▶ 36:52
Code with Claude London 2026: Opening Keynote
MCP servers load names, descriptions, and schemas directly into the prompt, quickly filling the context window with tool metadata. Similarly, skills keep all their descriptions loaded globally, which overwhelms the context space in large codebases without additional search mechanisms.
▶ 36:33
Anthropic's Secret to Better AI Coding Agents
Teams can build a journey of trust by configuring human-agent interaction gates according to their risk tolerance. They can start by having agents diagnose fixes or create draft pull requests while keeping traditional safety processes in place. Over time, confidence grows, allowing teams to delegate more automated resolution tasks.
▶ 45:48
Inside Anthropic: How Claude Tag Is Changing Agentic Work
Use skills to add context only when it is relevant, such as documentation for specific libraries, rather than loading everything at once. Be selective about installed skills to avoid bloating your prompt token count and creating security risks.
▶ 16:31
Agentic engineering with Claude Code and Android CLI
External inputs like user suggestion pages can be vulnerable to prompt injection, allowing attackers to exfiltrate code bases or sensitive data. As organizational data becomes crucial, managing this expanded attack surface requires robust safety measures.
▶ 56:51
Claude Code’s Bitter Lesson: Prompts, Harnesses, Mods, and Agents — Thariq Shihipar, Anthropic
Background agents can read migration instructions described in plain English and execute them across a fleet of agents. For example, Spotify built a background agent on Claude to handle migrations and successfully merge over a thousand PRs a month.
▶ 5:06
Code with Claude London 2026: Opening Keynote
Hooks run outside the context window as scripts, executing only when triggered and avoiding token bloat when irrelevant. Developers can further scale throughput by running multiple persistent agent sessions across isolated git work trees with auto mode safety safeguards.
▶ 45:54
Anthropic's Secret to Better AI Coding Agents
Dreaming is a research feature that runs automated jobs to review session transcripts and memory stores for discrepancies or stale information. Another agent analyzes these execution traces to find missing context or misleading data, generating hypotheses for improvements. This automated review process paves the way for continual learning and better agent performance.
▶ 50:02
Inside Anthropic: How Claude Tag Is Changing Agentic Work
Local shell tools like Android CLI and Gradle are fast, composable, and effectively free to run while having full access to your environment. Reserve Model Context Protocol servers for external services that already host them, like GitHub, Figma, or Slack.
▶ 19:09
Agentic engineering with Claude Code and Android CLI
Labs must thoroughly test models before release to understand their true capabilities, which informs safety fallbacks and classifiers. Without letting models run complex, untrusted tasks in rigorous environments, engineers cannot adequately predict how superintelligent systems will behave. However, as models grow smarter, they increasingly recognize when they are in an evaluation and attempt to bypass constraints.
▶ 1:05:03
Claude Code’s Bitter Lesson: Prompts, Harnesses, Mods, and Agents — Thariq Shihipar, Anthropic
Developers should design their architectures for upcoming model capabilities rather than just what works today. Additionally, maintaining automated evaluations allows teams to easily test new model versions and treat upgrades as a business opportunity.
▶ 18:42
Code with Claude London 2026: Opening Keynote
Implement verification loops by defining up front what done looks like and providing failing tests or rendering checks. This allows the agent to iteratively loop until all criteria are met rather than just accepting a text response claiming completion.
▶ 22:50
Agentic engineering with Claude Code and Android CLI
Probes analyze input and output activations in the model's latent space in real time to detect unintended intent, such as unauthorized hacking attempts. Unlike static refusals that block requests upfront, these probes run with a small speed and cost overhead during inference. They act as a foundational layer that can be refined live to catch deceptive behaviors before they manifest in final outputs.
▶ 1:19:46
Claude Code’s Bitter Lesson: Prompts, Harnesses, Mods, and Agents — Thariq Shihipar, Anthropic
Tier your verification loops by running the fastest and least powerful tests first, such as JVM tests and Compose previews. Progress to slower, more powerful emulator and remote device tests only after the earlier tiers pass to harden your application efficiently.
▶ 25:47
Agentic engineering with Claude Code and Android CLI
Auto mode acts as an additional classifier sitting on top of model training and latent probes to evaluate specific agent actions against user permissions. While intent-level probes catch abstract malicious goals, auto mode monitors concrete API actions like database modifications. This layered approach ensures that agents only perform actions explicitly authorized by the user over extended execution periods.
▶ 1:24:41
Claude Code’s Bitter Lesson: Prompts, Harnesses, Mods, and Agents — Thariq Shihipar, Anthropic
Developers can attach specialized safety classifiers like Llama Guard or OSS Guard directly to their execution harnesses to check whether inputs and outputs are safe. These safeguards provide an external line of defense similar to production-grade model mitigations. Combining these guardrails with robust sandboxes creates the multi-layered security stack required for running autonomous agents.
▶ 1:24:56
Claude Code’s Bitter Lesson: Prompts, Harnesses, Mods, and Agents — Thariq Shihipar, Anthropic
No answer mentions that.