Model Context Protocol

The new application layer belongs to agents We have reached a point where software engineering is facing its most significant disruption since the invention of the compiler. Malte Ubl, CTO of Vercel, argues that AI agents are not just a tool but a fundamental new kind of software. In the past, many automation ideas were discarded because they weren't economically viable using traditional coding methods. Writing a complex web of if-statements and hardcoded business logic was too expensive for niche tasks. Now, agents make that part of the software landscape viable. We are essentially speedrunning an experiment in economic elasticity: the cheaper it is to make software, the more software we will create. This shift moves the industry from a world of "builders vs. users" to a world where agents inhabit both roles. At Vercel, the team discovered that over 60% of their page views now come from agents, not humans. This has immediate implications for how we design interfaces. Graphical user interfaces (GUIs) are becoming a secondary concern, while command-line interfaces (CLIs) and APIs are the primary way software interacts with the world. When an agent is the user, it doesn't need a pretty dashboard; it needs a structured protocol. This paradigm shift forces us to rethink the very nature of software deployment and infrastructure, moving away from manual configurations toward automated, agent-driven environments. DeepMind pushes AI beyond the limits of language While language models dominate the conversation, Google DeepMind is expanding the boundaries of what intelligence can do in the physical and digital worlds. Raia Hadsell highlights that the future of AI isn't just about text—it's about understanding complex physical systems like the weather and creating immersive world models. GraphCast, a spherical graph neural network, can predict the state of the atmosphere 15 days out with higher accuracy than traditional physics-based models. In critical situations like Hurricane Lee, AI models provided accurate landfall predictions three days earlier than the industry's gold standard. Beyond weather, DeepMind is pioneering "world models" with Project Genie 3. This isn't just video generation; it is a playable, interactive environment created from a single image or text prompt. These models understand the "physics" of a scene—they know that if you walk into a puddle, the water should ripple. They maintain consistent memory, allowing a user to walk a mile in one direction and return to find the same structures intact. This represents a leap toward Artificial General Intelligence (AGI) that can navigate and interact with the world as humans do, opening new frontiers for education, simulation, and entertainment. Implementation is no longer a scarce resource The scarcity in software engineering has shifted. Ryan Lopopolo from OpenAI states a provocative truth: code is now free. In late 2025, with the release of advanced models like GPT-5.2, implementation ceased to be the bottleneck. An engineer no longer just manages their own output but acts as a staff engineer orchestrating dozens or thousands of agents simultaneously. The primary constraints are now GPU capacity and token budgets, not human typing speed. This abundance of code means that technical debt and refactoring are no longer terrifying burdens. If code is free to produce, it is also free to delete and rewrite from scratch. In this environment, the engineer's role moves toward systems thinking and delegation. We are entering the era of the "dark factory" for code, where vast repositories can be refactored overnight by swarms of agents. Vincent Koc describes running up to 15 parallel Cursor sessions to execute massive architectural changes that would have taken human teams months to complete. However, this velocity requires new guardrails. Instead of reviewing every line of code, engineers must build "harnesses"—automated systems of lints, tests, and security agents that enforce non-functional requirements. You don't review the code; you review the process that generated it. OpenClaw and the rise of personal AI infrastructure OpenClaw, created by Peter Steinberger, has become the fastest-growing project in GitHub history, signifying a massive demand for local, controllable AI. Unlike closed systems that require users to upload their sensitive data to a corporate cloud, OpenClaw allows individuals to run a powerful agent on their own infrastructure. This "hacker's way" of building AI provides a mechanism to bypass the silos created by big tech. If an agent can inhabit a local environment, it can access emails, files, and calendars without permission from a central authority. However, running an agent with the "keys to your life" introduces severe security challenges. Steinberger notes that OpenClaw is often targeted by security researchers because its power is inherently dangerous. Any system with access to private data and the ability to communicate can be exploited via prompt injection. The industry is responding with innovative deployment strategies. Sally Ann O'Malley of Red Hat advocates for running agents strictly in containers or Kubernetes pods to isolate secrets and ensure state recovery. By mounting specific directories and using secret references, developers can provide agents with the tools they need while maintaining a "blast radius" that protects the host system. Software fundamentals are the ultimate moat With agents churning out massive amounts of code, many fear that traditional engineering skills are becoming obsolete. Matt Pocock argues the exact opposite: software fundamentals matter more than ever. When you use AI to turn specs into code without understanding the underlying architecture, you often end up with "AI slop"—code that is brittle, shallow, and impossible to maintain. AI tends toward "software entropy," where every new change makes the system slightly more complex and disordered. To counter this, engineers must double down on Domain-Driven Design (DDD) and Test-Driven Development (TDD). By establishing a "ubiquitous language"—a shared terminology between the human and the AI—you ensure that the agent understands the domain it is working in. Furthermore, designing "deep modules" with simple interfaces allows an engineer to delegate implementation to the AI while keeping the overall system design clean. You treat the AI as a highly capable sergeant on the ground, but you remain the architect. If the codebase is well-structured, the AI performs brilliantly. If the codebase is a mess, the AI will only accelerate its collapse. Moving from tool calling to code execution The current standard for AI interaction, JSON-based tool calling, is reaching its limit. Sunil Pai from Cloudflare explains that stuffing hundreds of tool definitions into a context window is inefficient and slow. For large API surfaces—like Cloudflare's 2,600 endpoints—traditional Model Context Protocol (MCP) setups can consume millions of tokens. The solution is "Code Mode," where the model generates JavaScript that executes directly within a secure V8 isolate. This shift allows agents to inhabit the state machine of an application rather than just calling its functions. For example, an agent can inspect a canvas of strokes, recognize a game of tic-tac-toe, and execute a code snippet to draw a winning move without ever having been explicitly programmed for that game. This leads to the concept of "Generative UI," where software is no longer a static product but a dynamic environment that reconstructs itself for every user. The future of software is not a collection of buttons and forms; it is a safe sandbox where code does the talking, enabling a level of personalization and efficiency that was previously unimaginable.

Reclaiming Control Over AI Context Managing a complex development environment with Claude Code often leads to a "configuration sprawl" where global skills and local project plugins overlap. This clutter isn't just a mental burden; it directly impacts performance through context bloat. The Claude Code Organizer provides a centralized dashboard to visualize, move, and audit these assets. Prerequisites and Installation To use this tool, you need a working installation of the Claude CLI and Node.js. The organizer acts as a wrapper that reads your `.claude` directories. ```bash npx claude-code-organizer ``` Running this command launches a local web server, typically opening a dashboard in Google Chrome that maps out your Laravel Herd folders or any directory containing project-specific Claude configurations. Key Libraries & Tools * Claude Code: The primary CLI tool for AI-assisted coding. * Claude Code Organizer: A web-based management interface for skills and plugins. * MCP Servers: Specialized servers like Codex that extend the model's capabilities. * Visual Studio Code: Integrated for direct file editing from the dashboard. Managing Skills and Context Budgets One powerful feature is the ability to shift skills between scopes. If a specific prompt engineering skill is only relevant to a single repository, you can move it from global to local scope to prevent it from polluting other sessions. This directly affects your **Context Budget**. Every time you launch a session, Claude preloads configurations. The organizer calculates the token weight of these assets. For instance, four unused slash commands might consume 8,000 tokens before you even type your first prompt. Identifying these "heavy" skills (some exceeding 1.2MB) allows for surgical cleanup. Syntax and Practical Usage You can interact with the organizer directly within the terminal via the custom skill it installs: ```markdown /ccco # Launches the organizer dashboard from within Claude Code ``` This workflow allows you to audit `config.json` files and view Markdown documentation for installed plugins without manual directory navigation. Tips & Gotchas Always check the **Plan Mode** history within the dashboard. Claude Code saves project plans in hidden directories; the organizer makes these accessible for re-use or auditing. If your token usage feels high, prioritize removing legacy MCP Servers that you no longer actively use, as they contribute to the initial context payload.

The Architecture of Interactive Maps Mapping is no longer about static images. It is about a dynamic orchestration of data, rendering engines, and user interaction. To build effectively with Mapbox, you have to understand the fundamental relationship between the renderer and the style document. The renderer—whether it is Mapbox GL JS for web or the native SDKs for iOS and Android—acts as the engine. The style document is the blueprint. This style document is a JSON configuration that tells the renderer exactly what data to fetch and how to paint it on the screen. It defines layers, colors, and 3D properties. When a user zooms into London, the renderer is not just downloading a picture; it is requesting Vector Tiles. These tiles contain raw geographic data—coordinates for roads, buildings, and parks—which the client-side engine then draws in real-time. This architecture allows for fluid movement, 3D building extrusions, and the ability to change the entire look of a map instantly without reloading the page. Customization through Mapbox Standard and Studio The Mapbox Standard style serves as the premier base map. It is designed to be highly configurable while removing the heavy lifting of manual style management. For most developers, this is the starting line. It supports "lighting presets" that can shift a map from day to night or monochrome with a single parameter change. It also handles sophisticated features like 3D landmarks and detailed greenery automatically. However, when a generic base map is not enough, Mapbox Studio provides a professional design interface. Think of it as Photoshop for geographic data. Within Studio, you can import custom datasets, such as city-specific subway lines or proprietary business locations, and layer them precisely over the Mapbox base. One powerful technique involves zoom-dependent styling. You might represent data as simple circles when zoomed out to prevent clutter, then transition those points into detailed custom icons as the user zooms in. Once published from Studio, these styles are instantly accessible across all your applications via a unique style URL. Solving Search with Geocoding and Searchbox APIs Search is one of the most complex parts of any location-based app. Users expect smart, instant results, but the data behind addresses and points of interest (POI) are fundamentally different. Mapbox splits this functionality into two distinct tools: the Geocoding API and the Searchbox API. The Geocoding API is your workhorse for physical addresses. If you need to turn "123 Main St" into a coordinate, this is the tool. It is built for accuracy and permanent storage in administrative workflows. The Searchbox API, conversely, is built for the "fuzzy" nature of human intent. It includes POIs like restaurants, hotels, and landmarks. It uses a two-step process: "Suggest" and "Retrieve." As the user types, the Suggest endpoint provides a list of potential matches. Once the user clicks a result, the Retrieve endpoint fetches the full metadata—including things like wheelchair accessibility or specific building entrances. For web developers, the Mapbox Search JS library wraps these APIs into pre-built web components, handling the UI logic and network traffic so you can drop a search bar into your site in minutes. Navigation and Spatial Intelligence Navigation is more than just drawing a line between two points. It requires constant recalculation based on the user's live position and shifting traffic patterns. The Mapbox Navigation SDKs for mobile provide a "drop-in" UI that handles the entire turn-by-turn experience, including voice prompts and lane-level guidance. Under the hood, these SDKs communicate with the Directions API, which processes real-time traffic data to find the most efficient route. Beyond simple routing, Mapbox offers specialized spatial APIs like the Isochrone API and the Matrix API. An isochrone is a polygon representing the area reachable from a point within a certain time frame. This is a game-changer for delivery apps or real-estate platforms—instead of searching "within 5 miles," you search "within a 10-minute drive with current traffic." The Matrix API handles many-to-many calculations, allowing logistics platforms to determine the closest driver among a fleet of hundreds in a single request. Data Management and the Modern Developer Toolkit Getting data into the platform is often the biggest hurdle. The Data Workbench allows for direct, visual editing of geographic data in the browser. You can upload a GeoJSON file, realize an airport is missing, and draw the point manually using the editor tools. For larger, automated pipelines, the Mapbox Tiling Service (MTS) is the solution. It allows you to push raw data via an API, which then processes it into optimized vector tiles. This is essential for apps where data changes hourly, such as those tracking weather patterns or fleet movements. Mapbox is also pushing into the future of AI development with MCP (Model Context Protocol) servers. These tools allow AI agents to "understand" geography. By connecting an agent to a Mapbox MCP server, the AI can geocode addresses, generate static maps, or fetch directions on behalf of the user. This bridges the gap between large language models and the physical world, enabling agents that can help plan trips or analyze spatial trends through natural language. The Path Forward for Spatial Developers The ecosystem is vast, ranging from low-level tile access to high-level AI integrations. While the platform is robust, it relies on a community-driven feedback loop. Tools like the Contribute App allow developers and users to report road changes or speed limit updates, which eventually find their way back into the core data set. Whether you are building a simple store locator or a complex logistics engine, the key is to start small with the interactive playgrounds. Testing your API calls in a sandbox environment before writing a single line of production code is the best way to ensure your spatial logic is sound. Geography is messy, but the right toolkit makes it manageable.

The New Model on the Block Google recently launched Gemini 3.1 Pro within its Antigravity IDE, promising a significant leap in developer productivity. To see if the hype holds water, I put the model through a rigorous gauntlet: seven Laravel projects requiring complex API CRUD generation. While the integration feels seamless on the surface, the actual developer experience reveals a model still finding its footing in a competitive market. Performance and Latency Issues Speed defines the modern coding workflow. Unfortunately, Gemini 3.1 Pro lags behind. In side-by-side testing against Claude 3.5 Sonnet, Google's offering took six minutes to complete a task that Anthropic models finished in three. The model frequently pauses to calculate small details, launching internal help tools like "PHP design help" just to scaffold basic models. This suggests a lack of deep, native training on modern PHP frameworks. The Testing Gap and Agent Intelligence One glaring omission in the initial output was the lack of automated tests. While Gemini 3.1 Pro successfully generated models, factories, and controllers, it ignored the crucial step of verification. However, the model showed a flash of brilliance when prompted about this failure. It recognized its own "skills" via Laravel Boost and proactively corrected the mistake, eventually delivering 53 passing tests. This ability to discover and activate tools mid-stream is a clear positive, even if it requires manual intervention. Reliability and Quota Hurdles The Antigravity IDE experience remains plagued by stability issues. Random crashes and "terminated due to error" messages interrupted the workflow multiple times. Worse, the free tier quota is incredibly opaque. After only nine minutes of work on a Livewire project, the system cut off access entirely. Unlike the clear usage metrics provided by OpenAI, Google leaves developers guessing about how much "intelligence" they actually have left. Final Verdict: Catching Up Gemini 3.1 Pro is currently a secondary choice for heavy-duty Laravel development. It feels like a product in a "catching up" phase rather than a market leader. While the Gemini CLI shows promise for future MCP support, the current speed and reliability gaps make it hard to recommend over the more polished offerings from Anthropic.

Overview: The Shift to Agentic Development In the current software development landscape, we are moving beyond simple Large Language Models (LLM) wrappers toward sophisticated, autonomous entities known as AI agents. Unlike traditional chatbots that merely respond to prompts, these agents can use tools, access external data, and make decisions to execute complex business workflows. Redberry, a veteran Laravel partner, has formalized this process through LarAgent, an open-source tool designed to bring agentic capabilities directly into the PHP ecosystem. This approach matters because it allows developers to automate non-deterministic tasks—decisions that can't be hard-coded with simple if/else logic—while staying within a framework they already know and trust. Prerequisites To effectively build agentic systems with the tools discussed, you should have a solid grasp of the following: * **Modern PHP & Laravel**: Proficiency in service providers, configuration management, and the Laravel ecosystem. * **LLM Fundamentals**: Understanding of system prompts, temperature settings, and the difference between deterministic and non-deterministic outputs. * **API Integration**: Experience connecting with third-party services, as agents rely heavily on tool-calling to interact with the world. * **Vector Databases & RAG**: A basic understanding of Retrieval Augmented Generation (RAG) for providing agents with custom context. Key Libraries & Tools * **LarAgent**: An open-source package that provides the primitives for building agents in Laravel, including instruction management and tool-calling orchestration. * **Laravel AI SDK**: A first-party toolset from the Laravel team focused on standardizing AI interactions across different providers. * **MCP Client for Laravel**: A specialized package allowing Laravel applications to connect to Model Context Protocol (MCP) servers, giving agents access to an unlimited array of pre-built tools. * **Model Agnostic Layers**: Architectural patterns that allow switching between providers like OpenAI, Anthropic, or local models via configuration. The Anatomy of an AI Agent Sprint Building an agent isn't a linear coding task; it's a process of experimentation. A typical five-week proof of concept (PoC) focuses on time-boxing the non-deterministic nature of the project. Week 1: Discovery and Mapping Before writing code, you must map the business process. The goal is to identify which parts are deterministic (best handled by standard code) and which require an agent. If you can write a rule-based logic for a decision, you should. AI is reserved for the gaps where rules fail. Weeks 2-3: The First Prototype Using LarAgent, developers define the agent's instructions and the tools it can access. A "tool" in this context is often a PHP class or a specific API endpoint the agent can trigger. ```php // Defining a basic agent in LarAgent $agent = LarAgent::make('SupportBot') ->instructions('Assist users with order tracking.') ->tools([ OrderTrackingTool::class, InventoryCheckTool::class ]); ``` During this phase, you establish a benchmark data set. This is a collection of inputs and expected outcomes used to measure the agent's performance. Weeks 4-5: Iteration and Accuracy Initial success rates for agents often hover around 60-70%. The final weeks involve refining prompts, adjusting the orchestration of multiple agents, and tweaking tool definitions to push accuracy toward a production-ready 98%. This often involves "human-in-the-loop" design, ensuring a person reviews critical agent decisions. Syntax Notes & Orchestration Patterns One notable pattern in agentic development is the move away from a single, massive agent toward **multi-agent orchestration**. Instead of asking one agent to "manage an entire warehouse," you might have a "Receiver Agent," a "Stock Agent," and a "Dispatcher Agent." In LarAgent, this is handled through configuration-level model selection. Because different models excel at different tasks, you might use a smaller, faster model for simple categorization and a larger model for complex reasoning. ```php // Configuration-based model selection 'agents' => [ 'categorizer' => [ 'model' => 'gpt-4o-mini', 'temperature' => 0, ], 'analyzer' => [ 'model' => 'claude-3-5-sonnet', 'temperature' => 0.5, ], ] ``` Practical Examples * **Automated Test Case Generation**: Agents can scan project requirements and draft comprehensive test suites, which human developers then verify and approve. * **Legacy System Interfacing**: Using agents to interpret data from legacy systems that lack modern APIs, acting as a conversational or structured bridge between old and new tech. * **Regulated Industry Workflows**: In finance or healthcare, agents can pre-process documents and flag anomalies, significantly reducing manual labor while keeping a human as the final authority. Tips & Gotchas * **Avoid Tool Overload**: Exposing too many tools (more than 10) can overwhelm the LLM, leading to "hallucinations" or incorrect tool selection. Keep the agent's toolkit focused. * **Deterministic First**: Never use AI for something that can be solved with a simple database query or a standard function. It is more expensive and less reliable. * **Benchmark Early**: You cannot improve what you cannot measure. Build your test data set in week one so you have a baseline for every iteration. * **Legacy Blockers**: When integrating with ancient systems, expect blockers. Discovery should prioritize credential and API access to avoid stalling the sprint.

Overview: The Context Gap in AI Development AI agents have changed how we write code, but they often struggle with the nuances of specific frameworks. Standard models like Claude 3.5 Sonnet or GPT-4o possess vast general knowledge but lack the hyper-specific context of your local Laravel project. This lead to hallucinations, outdated syntax, or the AI suggesting patterns that conflict with your application's architecture. Laravel Boost solves this by acting as a bridge. It injects project-specific metadata, documentation, and "skills" directly into your AI agent's reasoning loop. Instead of manually feeding documentation to a chat window, Boost automates the context delivery. Version 2.0 introduces a major shift from a monolithic guideline approach to a modular, "skills-first" architecture. This reduces context bloat, saves on token costs, and makes the AI significantly more accurate by only providing the information it needs at that exact moment. Prerequisites To follow this guide and implement Boost 2.0, you should be comfortable with the following: * **PHP 8.2+:** Boost 2.0 has officially dropped support for PHP 8.1. * **Laravel 11 or 12:** Older versions like Laravel 10 are supported only by legacy versions of Boost (v1.x). * **Composer:** Basic knowledge of managing PHP dependencies. * **AI Coding Agents:** Familiarity with tools like Cursor, Claude Code, GitHub Copilot, or Juni. Key Libraries & Tools * **Laravel Boost:** The core CLI tool and package that manages AI context and skills. * **Laravel MCP:** A package for building Model Context Protocol servers, allowing AI agents to interact with your app's internal state (routes, database schemas, etc.). * **Remotion:** A React-based framework for programmatic video creation, often used as a demonstration of complex AI skill integration. * **Prism:** A Laravel package for working with LLMs, used to demonstrate how documentation can be bundled directly into vendor folders for AI consumption. Code Walkthrough: Installing and Configuring Boost 2.0 Setting up Boost 2.0 is a methodical process. It begins with the Laravel installer and moves into a randomized, aesthetically pleasing configuration CLI. 1. Installation First, ensure your Laravel installer is up to date to access the built-in Boost prompts during new project creation. If you are adding it to an existing project, use Composer: ```bash composer require laravel/boost --dev ``` 2. Initialization Run the install command to start the interactive configuration. ```bash php artisan boost:install ``` This command triggers a CLI interface featuring randomized gradients—a touch of "developer joy" added by Pushpak Chhajed. You will be prompted to select which features to configure: AI Guidelines, Agent Skills, or the MCP server. 3. Selecting Your AI Agent Boost 2.0 simplifies agent selection. Instead of choosing both an IDE and an agent, you now choose the specific agentic tool you use daily, such as Claude Code or Cursor. Boost will then automatically determine the correct file paths for these tools. 4. Automated Skill Syncing To ensure your AI context stays updated as your project evolves, add the update command to your `composer.json` file: ```json "scripts": { "post-update-cmd": [ "@php artisan boost:update" ] } ``` This ensures that every time you update your dependencies, Boost re-scans your `composer.json` and syncs the relevant skills for packages like Inertia, Tailwind CSS, or Livewire. Deep Dive into Skills vs. Guidelines Understanding the distinction between these two features is critical for a clean development workflow. Guidelines: The Global Rules Guidelines are persistent. They contain high-level rules that the AI should *always* know. For example, if you always use Pest for testing or strictly follow an Action-based architecture, these belong in your guidelines. However, shoving every package's documentation into a guideline leads to "context fatigue," where the AI becomes overwhelmed and starts to hallucinate. Skills: The On-Demand Context Skills are modular Markdown files. They aren't loaded into the AI's memory until they are needed. Each skill has a name and a description in its front matter. When you ask the AI to "build a new UI component with Tailwind," the agent sees the keyword "Tailwind," looks at its available skills, and activates the Tailwind CSS skill. This keeps the prompt lean and the output precise. Syntax Notes: Custom Skill Creation Creating a custom skill allows you to automate highly specific tasks, like generating pull request descriptions or adhering to internal API versioning standards. Skills rely on a specific Markdown front matter format. ```markdown --- name: my-custom-skill description: Use this skill when generating API endpoints or PR descriptions. --- My Custom Skill Rules - Always use the `App\Actions` namespace for business logic. - Ensure all API responses are wrapped in a standard `JsonResource`. - Pull Request descriptions must include a 'Breaking Changes' section. ``` When you save this in a local `.boost/skills` directory and run `php artisan boost:update`, Boost replicates this file into the hidden configuration folders of your chosen AI agents (e.g., `.cursor/rules` or `.claudecode/skills`). Practical Examples Automating Pull Requests You can create a skill that teaches an agent how to use the GitHub CLI. By invoking the skill with a slash command (e.g., `/create-pr`), the AI can analyze your staged changes, write a formatted description, and execute the CLI command to open the PR. Package-Specific Intelligence If you build a project using Filament, you don't want the AI thinking about Filament when you are just debugging a console command. By using a Filament skill, the AI only accesses those specific layout and component rules when you are actively working on the admin panel. Tips & Gotchas * **Git Management:** Never commit the auto-generated agent folders (like `.cursor/rules`) to your repository. These are local mirrors. Only commit the `.boost` folder and your `boost.json` file. This allows your teammates to run `boost:install` and get the exact same AI behavior on their machines. * **Hallucination Prevention:** If your AI starts ignoring your project structure, check your guideline length. If it exceeds 500 lines, move package-specific rules into individual skills. * **Legacy Projects:** Do not attempt to use Boost 2.0 on Laravel 10 projects. The dependency tree for the new MCP features and skills requires the modern internals found in Laravel 11 and up. * **Manual Invocation:** If an agent fails to auto-detect a skill, you can usually force it by using a slash command in the chat interface. Most modern agents support `/` to list and select active skills.

The Hidden Cost of Context Pollution Every developer working with Claude Code eventually hits the same wall: the dreaded context limit warning. It’s not just an annoyance; it’s a productivity killer. When your context window fills up with junk data, the model's reasoning degrades, and your costs—or at least your token usage—skyrocket. Managing this space requires a shift in how we think about documentation and tool interaction. By treating context as a finite, precious resource, you can maintain model sharp-sightedness even in complex projects. Filter Your Database Schema Requests One of the silent killers of context space is the Model Context Protocol (MCP) toolset. During recent experiments with the Laravel Boost MCP, a bug revealed that the tool was pulling far more data than necessary, bloating the context by thousands of tokens just to understand a simple table structure. You don't have to wait for a bug fix to reclaim this space. You can proactively override tool behavior by adding specific instructions to your `claude.md` file or directly in your prompts. By explicitly telling the model to "filter only the current database" or narrow its scope to specific tables, you can slash token usage from 15,000 down to a mere 0.5% of your total context. This surgical approach ensures the agent sees the schema it needs without drowning in metadata. Slice Your Documentation for Maximum Speed We often create massive, 1,000-line "Project Phase" documents to ensure nothing gets missed. However, referencing a single phase within a giant file still forces Claude Code to ingest the entire document. This results in massive context pollution because the model reads everything before it can filter for its specific task. The fix is simple but transformative: **slice your docs**. Instead of one monolithic file, break your roadmap into individual Markdown files—one for each phase. Transitioning from a 1,000-line master file to a 160-line phase-specific file can reduce a user message’s context footprint from a heavy burden to a negligible 1%. The Power of Post-Prompt Analysis Efficiency isn't a one-time setup; it's a habit. Claude Code has the built-in capability to analyze its own token usage and list the processes or tools that consumed the most resources. After completing a task, ask the agent to list what actually ate the context. This reveals patterns—like a test file that’s unexpectedly large or a system prompt that’s redundant—allowing you to refine your workflow for the next prompt. Conclusion Optimizing your AI workflow is about more than just writing better code; it’s about managing the environment where that code is generated. By filtering your database queries and modularizing your documentation, you ensure that the AI stays focused on the logic that matters. Start auditing your token usage today and see how much faster your development cycle becomes when you cut the dead weight.