I'm writing this because something important has changed and I don't think most people have fully clocked it yet. Claude is currently the best LLM going around for 'work stuff', but the bigger change is what it's done to connectivity. You no longer need to lean so heavily on apps, UIs and dashboards, or understand APIs and endpoints, to make systems talk to each other. You tell Claude what you want done, in plain English, and it handles the plumbing behind the scenes. That changes who can build things, and how fast.
Most of you have heard of Claude Code by now. Some of you have used it. A few of you are starting to build real things with it. But I don't think enough people understand what that means in practice.
Everything is code. You are now a coder*.#
Claude doesn't just answer questions. It writes, runs, debugs and ships working code from a plain English description of what you want. Need a script that pulls data from your CRM, cross-references it against a spreadsheet and flags anomalies? Describe it. Need a prototype dashboard for a partner meeting on Thursday? Describe it. Need to automate the repetitive part of your week that eats three hours every Monday morning? Describe it. Turn it into a skill.
The distance between "I have an idea" and "I have something that works" is now very short.
*To be clear, this doesn't make everyone a software engineer. Building production-grade systems, managing architecture at scale, reasoning about security and infrastructure: that's still a craft, and it still takes years to develop. What's changed is access. You no longer need to know how to code to get useful things built. Whether you sit in business development, marketing, ops, finance or legal, you can now build things that used to need a ticket, a sprint and a six-week wait. That changes what's possible in your role.
Your ability to communicate is now your most valuable technical skill.#
The main interface for this technology is English. As Andrej Karpathy put it, "the hottest new programming language is English."
How clearly you can say what you want (break a problem down, describe the outcome, set the constraints) decides whether you get a mediocre result or a good one. The same tool in two people's hands will produce very different output.
If you can write a sharp brief, you can build software. If you can explain a process to a new starter, you can automate it.
Claude and Claude Code are different tools.#
They're built on the same foundations, but they work very differently and you use them for different jobs.
Claude can be just a chat interface. You go to claude.ai, type a question or a request, and it responds. It's very good for drafting documents, brainstorming, analysing text, summarising information and thinking through problems with you. If your task lives entirely inside a conversation (writing, thinking, advising), Claude is the right tool. Most of you already know this one.
Claude Code runs in your terminal (the command line on your computer) and works inside a folder on your machine. The folder part matters. When you launch Claude Code, you point it at a directory, and it can read, write, create and run files inside it. So as well as talking, it acts: writing scripts, running code, processing data, building tools. It can also connect to external systems through MCP (Slack, Google Drive, databases, APIs) and coordinate work across them.
The simplest way to think about it: Claude is for conversations, and Claude Code is for building and doing. If you want help writing an email, use Claude. If you want to automate your weekly reporting, build a dashboard or process a dataset, use Claude Code.
You don't need to be technical to use Claude Code, but you do need to understand that it works inside a folder on your machine. When you start a session, you open your terminal, go to the folder you want to work in (or create one) and launch Claude Code from there. Everything it builds lives in that folder. If that sounds unfamiliar, it takes about five minutes to learn.
What's different under the hood.#
Four things set Claude Code apart from a chatbot, and none of them need a technical background to follow.
It works in a loop.#
Many AI coding tools sit inside a text editor and offer autocomplete. You're still driving, and the tool is finishing your sentences. Claude Code works differently. You give it an objective, and it runs a loop: it thinks about the problem, takes an action, looks at the result and adjusts. If something breaks, it reads the error, works out the cause and fixes it without you touching anything.
It's closer to delegating to a colleague than using spellcheck. You define what "done" looks like, and it works out how to get there.
It handles large data by writing code.#
When you throw a lot of information at most AI tools (thousands of spreadsheet rows, a massive log file), the quality of their reasoning falls off a cliff. They're trying to hold too much in their head at once.
When Claude Code hits a big data problem, it writes its own script to process the data, runs it and brings back only the summary it needs. Instead of reading every line of a 10,000-row spreadsheet to answer your question, it builds a calculator, runs the numbers and gives you the answer.
It connects to the tools you already use.#
Through the Model Context Protocol (MCP), Claude Code connects to Google Drive, Slack, project trackers and internal databases. You can point it at a product spec in a doc, and it'll pull context from Slack threads, check related tickets and start working across those systems.
It breaks big jobs into smaller ones.#
Long, complex tasks are where most AI falls apart. The model forgets what it was doing, drifts off course or runs out of working memory halfway through. Claude Code breaks big jobs into smaller, independent pieces, each handled by a separate sub-agent that starts with a clean slate. It also saves its state before every major change, like a checkpoint in a game, so if something goes sideways you roll back to the last good state instead of starting over.
Where it falls short.#
It isn't magic, and it has real failure modes. Knowing them is part of using it well.
It over-engineers. Left alone, Claude will often build something far bigger than you asked for. It's optimised to generate output, which gives it a bias toward adding complexity. Be specific about what you want, and just as specific about what you don't.
It can spin its wheels. If it hits a problem it can't crack, it won't always stop and tell you. It can get stuck trying the same failing approach over and over, so set clear boundaries and check in on longer tasks.
It needs proper context. The more clearly you define the problem, the constraints and what success looks like, the better the output. Vague inputs get vague results, which is why the communication skill from earlier matters so much.
Your job moves from writing the steps yourself to setting the guardrails and defining what good looks like.
The gap is opening fast.#
Using tools like Claude Code doesn't feel optional to me anymore. The gap between people who work these tools into their day and people who don't is widening quickly. Every week you're not using Claude (or an equivalent) in your real work, you're leaving capacity on the table.
You don't need permission or a training course to start. Open it, bring a real problem from your week and work through it in plain English. Expect to get it wrong a few times before it gets better.
Views expressed here are my own and don't represent any other parties.