We have spent this series building from the ground up.
How the internet works. What LLMs are. How AI agents operate. What RAG does. How MCP connects everything. How to build with AI regardless of your role or technical background.
This article is where the series gets practical.
Anthropic has released two tools that sit at the centre of how the most effective builders, engineers, PMs, operators, and enterprises are working with AI today. Claude Code and Claude Cowork.
Most people have heard of them. Far fewer understand the full surface area of what they can actually do.
This article is the complete picture. We cover every major feature, how each one works, who it is built for, and how different roles and organisations can put them to use right now.
Part One: Claude Code
What Claude Code actually is
Claude Code is not a chatbot with a code block. It is not autocomplete. It is not a coding assistant that suggests the next line.
Claude Code is an agentic coding system. It runs in your terminal, reads your entire codebase, writes changes across files, runs commands, executes tests, and delivers committed code. It acts toward a goal rather than responding to one prompt at a time.
The distinction is fundamental. Most AI coding tools respond to you. Claude Code acts for you.
When you brief it to build a feature, it opens the relevant files, writes the implementation, runs the tests, fixes the failures, and tells you what it built and why. It is the difference between a recipe and having someone cook the meal.
Anthropic reports that the majority of code at Anthropic is now written by Claude Code. Stripe deployed it across 1,370 engineers and one team completed a 10,000-line Scala-to-Java migration in four days, work that was estimated at ten engineer-weeks. Ramp cut incident investigation time by 80%. Rakuten reduced average feature delivery from 24 working days to 5.
Feature 1: Full codebase awareness
Claude Code reads your entire project, not just the file you are looking at.
It searches directories, traces dependencies, and understands how modules connect. It knows which files import which, where configuration lives, what patterns the codebase follows, and how the API is structured.
This means when you ask it to extend something it extends it consistently with what already exists. It does not introduce patterns that conflict with your conventions or duplicate logic that already lives somewhere else.
How engineers use this. Onboarding to a new codebase takes minutes instead of weeks. Ask Claude Code to walk you through how a part of the system works and you get a complete, accurate explanation from the actual code.
How PMs use this. Before writing a spec, ask Claude Code to explain what a specific part of the product currently does. You get a plain-English walkthrough of the real implementation, not what someone documented six months ago.
How enterprises use this. New engineer onboarding is compressed dramatically. Senior engineers stop being pulled away to explain architecture. The deep knowledge of how a system works becomes accessible to the whole team.
Feature 2: File reading, editing, and creation across the project
Claude Code can open any file, read it, edit it, and create new files. It makes the changes, it does not just suggest them.
You can ask it to refactor a function across every file it appears in. To rename a variable project-wide. To restructure a directory. To create a new module following the pattern of an existing one. To migrate deprecated patterns across an entire repository.
How engineers use this. Large refactors that previously required days of careful manual work can be executed in hours. The consistency of AI-driven changes across thousands of files eliminates the class of errors that creep in when humans refactor at scale.
How analysts use this. Create data processing scripts, transform exports, generate reports. Claude Code writes and runs the script in the same conversation.
How enterprises use this. Wiz migrated a 50,000-line Python library to Go in roughly 20 hours of active development, a project estimated at two to three months manually. Enforcing new coding standards across large repositories becomes a managed process rather than an impossible ask.
Feature 3: Command execution
Claude Code runs commands in your terminal. Not just suggests them. Runs them.
It installs dependencies, runs tests, starts development servers, executes migrations, runs builds, and runs custom scripts. It reads the output and adjusts its approach if something fails. When tests fail it reads the errors, fixes the code, and runs the suite again until everything passes.
How engineers use this. Ask it to run CI, identify failing tests, and fix them. It monitors pipelines and commits fixes automatically. The feedback loop that previously required manual intervention runs end to end.
How founders use this. Set up a local development environment, scaffold a prototype, and get to a working state without needing to follow setup documentation step by step.
How enterprises use this. Automate deployment sequences, database migrations, and environment setup. Build internal tooling that runs scheduled operations using Claude Code as the orchestration layer.
Feature 4: Multi-step agentic task execution
Claude Code plans and executes multi-step tasks without you managing each step.
You can ask it to build an entire feature end to end. It will write the migration, build the API endpoint, add the service layer, write the tests, update the documentation, and report when it is done. Each step is visible. You can intervene, redirect, or refine at any point.
Developers control how much autonomy Claude Code has, from approving every action to letting its built-in classifiers distinguish safe actions from risky ones automatically.
How engineers use this. Run multiple Claude Code sessions in parallel, delegating different tasks across the codebase simultaneously. The throughput of a single engineer multiplies significantly.
How PMs and founders use this. Build a complete working prototype from a brief. Give it the context, the stack, and the requirements. It builds. You review. You ship.
Feature 5: Git integration
Claude Code understands Git natively. It reads commit history, understands what changed and when, creates commits, and works within branching structures without you remembering the syntax.
Developers do not need to memorise commands for Git, Kubernetes, or other CLI tools. They describe what they want and Claude Code uses tools like the GitHub CLI natively, running the right commands with the right syntax.
How engineers use this. Ask it to review the diff before a PR goes out. It reads what changed, flags anything that should not be there, and gives you a pre-review before the team sees it. Ask it to write an accurate commit message that describes what actually changed.
How PMs use this. Ask Claude Code to summarise what changed in the last sprint from the commit history. You get a plain-English changelog without reading diffs.
How enterprises use this. Automate changelog generation, release notes, and audit trails. CI pipelines on GitHub and GitLab can trigger Claude Code to inspect failures and commit fixes automatically.
Feature 6: MCP tool integration
Claude Code connects to MCP servers. This means it can reach beyond your local codebase to any system you have connected.
We covered how MCP works in Part 5 of this series. Claude Code is where that infrastructure pays off most visibly. Connect it to your database and it can query production data to inform a fix. Connect it to your monitoring system and it can read error rates while debugging. Connect it to your internal documentation and it can reference your standards while writing code.
How enterprises use this. Claude Code becomes the interface to the entire internal toolchain. Engineering, data, infrastructure, and documentation all become accessible through a single agent. Non-engineering teams at Ramp now query their data warehouse in natural language instead of writing SQL because of this kind of integration.
Feature 7: Routines and scheduled tasks
Claude Code can run scheduled tasks on Anthropic-managed infrastructure. Routines keep running even when your computer is off. They can also trigger on API calls or GitHub events.
You create them from the web, the desktop app, or by running the schedule command in the CLI. Desktop scheduled tasks run on your machine with direct access to your local files and tools.
How engineers use this. Automate recurring checks, scheduled migrations, and periodic code quality scans. Set it to monitor a repository and flag drift from your standards automatically.
How enterprises use this. Build operational automations that run on a schedule independently of any person being present. The infrastructure for reliable, repeatable engineering operations.
Feature 8: Remote control via Dispatch
Claude Code sessions are not tied to a single surface. You can move work between environments as your context changes.
Start a task from your phone and pull it into your terminal. Kick off a long-running task on the web or iOS app, then continue it in your terminal. Hand off a terminal session to the desktop app for visual diff review.
Development tasks route through Claude Code and knowledge work tasks route through Cowork, with results returning as a pull request, spreadsheet, memo, or whatever output the task requires.
How engineers use this. Dispatch a task from your phone on the way to a meeting. Return to your desk to find it completed and ready to review.
How enterprises use this. Unblocked development that does not require someone to be at their desk. Long-running tasks continue independently.
Feature 9: VS Code and Cursor integration
Claude Code has a VS Code extension that provides inline diffs, mentions, plan review, and conversation history directly in your editor. You can also install it for Cursor.
This brings the full agentic capability into the editor environment without requiring you to switch to a terminal. You can reference specific files and functions, see proposed changes as diffs before they are applied, and keep a history of what the agent has done in the current session.
How engineers use this. Stay in your preferred editor while having full Claude Code capability available. See changes as diffs and approve or reject them without leaving the file you are working in.
Feature 10: Extended reasoning for complex problems
For complex tasks, Claude Code engages extended reasoning. It thinks through the problem before acting.
This matters for architecture decisions, complex debugging, and multi-system design. The reasoning is visible and auditable, not just the conclusion.
How engineers and architects use this. When facing a genuinely complex decision, ask Claude Code to think through the options before recommending one. You see the trade-off analysis, not just the answer.
Part Two: Claude Cowork
Claude Cowork is the same agentic architecture as Claude Code, wrapped in the familiar Claude desktop interface for non-technical users.
Where Claude Code is built for the terminal and the codebase, Cowork is built for knowledge workers. For the people who live in documents, spreadsheets, folders, and files. No terminal required.
You find Cowork in your desktop app alongside Chat and Code. Switch to Cowork when you want Claude to execute non-coding tasks rather than just respond to them.
Anthropic built the entire Cowork product using Claude Code in approximately two weeks. The majority of Cowork usage comes from outside engineering teams, from operations, marketing, finance, and legal, teams handling the work that surrounds their most critical tasks: project updates, collaboration decks, research sprints.
Feature 1: File system access and management
In regular chat, Claude responds to your messages but cannot access your files directly. In Cowork, Claude has permission to read, edit, and create files in folders you specify.
You choose which folders Claude can access. Before it acts, it shows you the plan and waits for your approval. You can redirect, refine, or take a different approach at any step.
Point it at a folder of drafts, downloads, and attachments and ask it to rename, sort, deduplicate, or surface what is relevant. File systems accumulate faster than anyone can organise them. Cowork handles the maintenance that nobody has time for.
How operators use this. Automate file management at the end of every project. Organising client deliverables, archiving completed work, preparing handoff packages. The overhead that accumulates around organised work runs automatically.
How analysts use this. Process batches of exported reports, organise datasets by date or source, prepare data files for analysis without spending time on the organisational overhead.
How enterprises use this. Standardise how files are named, organised, and archived across teams. What previously depended on individual discipline becomes a consistent, automated workflow. Anthropic streams file access through OpenTelemetry to SIEM systems for enterprise audit compliance.
Feature 2: Document reading and synthesis
Cowork can open and read documents across formats. PDFs, Word documents, spreadsheets, presentations. It reads them and produces structured outputs from them.
The hardest part of writing a report is rarely the writing. Hand off a set of source files and Cowork produces a structured draft, handling assembly and synthesis so the work that remains is refinement.
Ask it to summarise a 50-page report in three paragraphs. Pull all action items from a set of meeting notes. Compare two contract versions and flag the differences. Extract specific data points from a folder of financial reports.
How PMs use this. Drop a competitor update or a lengthy research report into Cowork and ask for the three things that matter for your roadmap. What previously took an afternoon takes five minutes.
How legal teams use this. Review contract batches, flag non-standard clauses, and produce plain-language summaries for stakeholders without reading every word manually.
How enterprises use this. Process large volumes of inbound documents at the first layer before anything reaches a human. Customer feedback, partner documents, regulatory filings. Tedious tasks that previously got skipped, like scanning all feedback or reviewing every submission, now get done, leading to better decisions.
Feature 3: Skills for native file format handling
Cowork includes Skills that give it native handling for office file formats. Excel, PowerPoint, Word, and PDF.
The PDF skill goes beyond basic reading to support merging, splitting, and form-filling. The Excel and Word skills produce formatted outputs that look like something a professional produced, not something generated by an AI.
You can also build custom Skills via the Skills API to package your own domain expertise and organisational workflows, making Cowork understand your specific context and output formats.
How finance teams use this. Ask Cowork to analyse a spreadsheet of operating expenses and produce a formatted Word report with an executive summary, category breakdown, and month-over-month trends. The assembly and formatting run automatically.
How enterprises use this. Package your standard report templates, naming conventions, and output formats as a custom Skill. Every output follows your standards automatically without manual reformatting.
Feature 4: Connectors and integrations
Cowork connects to external services through connectors. Slack, Chrome, Gmail, Google Drive, AWS, n8n, Honeycomb, Fellow.ai, and hundreds more are available in the connector catalogue.
In Cowork, connectors gain a new dimension: filesystem access. A connector that pulls data from an external service can now save that data locally, or use local files as input for external actions.
Claude picks the fastest path for any task: a connector for direct integrations like Slack, Chrome for web research, or your screen to open apps when there is no direct integration. Using a direct integration is faster and more precise than navigating through the screen, so the screen is a last resort.
How operations teams use this. Connect Slack and have Cowork summarise all relevant channel activity into a daily briefing. Connect your calendar and have it prepare a meeting brief from local files before your first appointment.
How enterprises use this. Manage connector access with role-based controls. The enterprise connector catalogue includes tools for cloud resources, observability, workflow automation, and meeting insights. Each connector is Anthropic-reviewed before it enters the catalogue.
Feature 5: Multi-step workflow automation
Cowork executes multi-step workflows on your behalf. You describe the outcome. It figures out how to get there.
For recurring tasks, you tell Claude once and it handles the rest. The consistency of an automated workflow is valuable independently of the time it saves.
Process a folder of files and produce a summary for each. Read a set of reports and create a consolidated output. Take a list of documents, extract specific information from each, and produce a structured comparison table. Package a pitch deck as PDF and attach it to a meeting invite.
How founders use this. Automate the administrative overhead that accumulates around a growing company. Processing applications, reviewing submissions, organising inbound information.
How HR teams use this. Process job applications, organise candidate materials, and prepare structured summaries before interviews. What used to require a dedicated coordinator can run automatically.
How enterprises use this. Build repeatable workflows for the operational tasks that happen every week. Report compilation, data aggregation, document routing. Enterprise deployment includes management controls to track Cowork usage and spend across the organisation.
Feature 6: Computer use
When there is no direct connector for a service, Cowork can interact directly with your screen. It opens apps, navigates your browser, fills in forms, and operates your computer the way a human would.
Computer use is in research preview for both Cowork and Claude Code. When Claude does interact with your screen, it asks permission before accessing each application. You see what it is doing and can stop it at any step.
Claude reaches for connectors and integrations first, falls back to the browser when needed, and only uses your screen as a last resort. This matters because direct integration is faster and more precise than screen navigation.
How non-technical teams use this. Access services and workflows that have no direct API or connector. Any application you can open on your desktop, Cowork can operate on your behalf.
Feature 7: Dispatch, remote control from your phone
Dispatch is a workflow layer inside Cowork that connects the Claude mobile app to the Claude desktop app.
You send a task from your phone. Claude routes it to the appropriate desktop session, runs the work on your local machine with access to your files, connectors, and applications, and sends you the result.
A Dispatch thread does not reset, so Claude retains context from previous tasks. You can start on your phone, continue on your desktop, and the session remembers everything.
Practical uses. Prepare for a meeting when running late. Export a pitch deck as PDF and attach it to a calendar invite. Find a file and summarise it before you arrive at your desk. Compare three contracts by price and renewal terms. Process receipts for reimbursement. Set up a daily briefing that checks your calendar, email, and Slack before your first meeting.
How enterprises use this. Unblocked knowledge work that does not require someone to be at their desk. Asynchronous task delegation that routes to the right tool, Code for development tasks, Cowork for everything else.
Part Three: Who uses what and how
The product manager
PMs use Cowork for the surrounding work that pulls them away from decisions. Research synthesis. Stakeholder updates. Document organisation. Meeting preparation. The administrative weight of the role shrinks.
They use Claude Code to understand what is actually built versus what was specified. To generate prototypes before committing engineering resources. To build internal tools the team needs but engineering cannot prioritise.
The combination means arriving at every conversation better informed and every sprint better directed.
The software engineer
Claude Code is the primary tool. Full codebase awareness, multi-step execution, test writing, debugging, Git integration, and CI monitoring cover the mechanical parts of the engineering workflow.
The leverage is significant. Multiple parallel sessions, each delegated a different task across the codebase. The throughput of a team from a single engineer.
Cowork handles the non-code engineering work. Technical spec documents, post-mortem drafts, architecture decision records. The writing that engineers do but rarely enjoy.
The product designer
Cowork handles the surrounding work. Research synthesis. Brief generation. Handoff documentation. Stakeholder updates. Competitive analysis from a folder of screenshots and PDFs.
Claude Code becomes useful as designers engage with the technical layer. Understanding what is actually implemented versus what was designed. Generating accurate handoff annotations from the real code. Bridging the gap between design intent and engineering reality.
The analyst
For analysts, Cowork is transformative. Processing large volumes of data files, documents, and reports is exactly what it is built for.
Batch summarisation of research. Extraction of structured data from unstructured reports. Consolidation of multiple sources into a single output with the Excel and PDF skills doing the heavy formatting. Tedious tasks that previously got skipped now get done routinely.
Claude Code is available for analysts who want to go further. Writing and running data processing scripts, building internal dashboards, automating report generation pipelines.
The founder
Founders have the broadest set of tasks and the least time for any of them.
Cowork handles the operational layer. Document processing, communications drafting, information organisation, and all the administrative overhead that accumulates around a building company.
Claude Code handles the building layer. Prototypes before engineering begins. Internal tools the team needs. Automations that remove recurring manual work. A kind of leverage that was not accessible before.
The enterprise
Enterprises have different opportunities and different requirements.
At scale, consistency matters as much as speed. Cowork standardises how documents are processed and written outputs are produced. Claude Code standardises how engineering workflows execute. The floor rises across the entire organisation, not just for the strongest individuals.
The governance layer matters. Cowork now includes role-based access controls, granular MCP permissions, and OpenTelemetry streaming to SIEM systems. Claude Code works locally, meaning your codebase does not leave your environment. Cowork works on your desktop, meaning your documents stay on your infrastructure.
The enterprise deployment controls for Cowork let you manage feature access, control spend, and track usage across the organisation. These are not afterthoughts. They are design decisions that came from the enterprise rollout feedback that followed the initial launch.
Part Four: How to start today
Starting with Claude Code
Claude Code is available through the Anthropic API, directly in the Claude interface, via a VS Code extension, and for Cursor. Enterprise users can run it through Amazon Bedrock or Google Cloud Vertex AI.
Start with a real task, not a test task.
If you are an engineer: Brief Claude Code on a current bug. Give it the error, the symptom, and the relevant file. Let it trace through the codebase and propose a fix. Watch how it works before you approve anything.
If you are a PM: Ask Claude Code to explain what a specific part of your product does by reading the relevant files. You will understand your own product more clearly in twenty minutes than from any sprint review.
If you are a founder: Give it your tech stack and ask it to scaffold the feature you have been thinking about. Use the system prompt structure from earlier in this series to brief it well.
Starting with Cowork
Cowork is available on all paid Claude plans through the Claude desktop app on Mac and Windows. Find it in your desktop app alongside Chat and Code.
Start with the task that takes the most time and requires the least judgement. A document review. A folder organisation. A report summary from a set of source files. The tasks where the work is real but the thinking is mechanical.
First task to try: Point Cowork at a folder of documents and ask it to produce a structured one-page summary of each. See what it produces, how it organises the output, and where you would refine the brief for next time.
Second task: Give it a recurring task you do every week. A weekly status compilation. A report assembly. A meeting prep brief. Ask it to set this up as a scheduled routine so it runs automatically going forward.
Reflection for the week
We started this series with something simple. The internet is just two things talking to each other.
Every article since then has added one layer. LLMs. Agents. RAG. MCP. APIs. Fundamentals. The empowered PM. How to build with AI regardless of your role.
Claude Code and Cowork are where all of that lands.
They are not abstract concepts. They are tools you can open today. Features you can configure this afternoon. Tasks you can delegate before the end of this week.
The people who understand what these tools are doing underneath, the ones who have followed this series from the beginning, are the ones who will use them most effectively. Not because they read a feature list, but because they understand the systems the features are built on.
That is what this entire series has been about.
The question worth sitting with this week:
Which one task in your role could you hand to Claude Code or Cowork today? And if that went well, what would you hand it next?
Want to build this foundation properly?
Everything in this series, from how the internet works to how to use Claude Code and Cowork effectively, is what we teach in our 14-week Amakora AI Product and Systems Fellowship starting next month.
Live classes. Real projects. A cohort of people building alongside you.
Apply here: amakoragroup.com/apply
Until next week,
Tochii
Founder, Learn with Tochii
Inspire · Educate · Empower
📧 contact@tochukwuachebe.com
🌐 https://www.tochukwuachebe.com