Chat, app, agent, coding: what these terms actually mean

The four ways you can use AI right now, explained in plain English for founders who've heard the buzzwords but want to know what they actually mean.

If you've spent any time around the AI conversation in the last two years, you've heard a lot of words.

Chat. App. Agent. Copilot. Coding. Vibe coding. Models. Tokens. Frontier. Embedded.

Most of these words are becoming second nature, but what do they actually mean?

If you are feeling lost, you're not alone. Every small business owner I've spoken to in the last six months has admitted, sooner or later, that they don't really understand half the terminology being thrown around. They've assumed everyone else does. They don't.

I have broken these down into four categories which form a framework. Once you understand them, every AI tool you encounter will fit into one of them, or be a combination of two. The categories themselves do not change. The specific tools inside them change every six months, which is why building using this framework is more valuable than learning any specific tool.

Why this matters more than people realise

Most of the bad decisions small business owners make about AI come from terminology confusion.

Someone sells you an "AI agent" that turns out to be a glorified chatbot. Someone tells you to "use Copilot" without specifying whether they mean Microsoft Copilot, GitHub Copilot, or one of the dozen other things calling themselves Copilot. Someone explains that you need to "build with AI" without distinguishing between using a chat tool well, integrating an AI feature, or actually coding something custom.

These are not trivial distinctions. They have different costs, different use cases, different levels of maturity, and different verdicts on whether you should pay attention right now.

Here's the framework.

The four categories we will explore are:

1. Chat: a conversational AI tool
2. Apps: AI features built into apps you are already using
3. Agents: AI that can take actions on your behalf
4. Coding / building: AI that builds software for you

Let's dive in.

Category 1: Chat

What it is, in plain terms

Chat is a conversational AI tool. You type something, it responds. You can have a back and forth conversation with it. You can ask it to do something, give it feedback, ask it to try again, ask it to explain. It works like a smart colleague who's read most of the internet and is endlessly patient with your questions.

This is the category most people have already encountered, even if they did not realise it. If you've ever used ChatGPT, you've used a chat tool.

The five tools worth knowing

• ChatGPT, made by OpenAI: strong all rounder, widely known, large integration ecosystem
• Claude, made by Anthropic: strong writing quality, good with long documents, more careful reasoning
• Gemini, made by Google: tight Google Workspace integration and strong live-information workflows
• Copilot, made by Microsoft: tight Microsoft 365 integration and often included in business plans
• Grok, made by xAI: the same company behind X, and it has some great tools and access to live posts

The free versions of all five are genuinely usable. Copilot is mostly paid, but some free versions exist. The paid versions are better: more capable models, longer conversations, file uploads, and better tools.

For a small business owner starting today, I'd recommend Claude or ChatGPT. Both are excellent and there's no meaningfully wrong choice between them. Pick one, use it for two weeks, then decide if you want to switch. Most people end up using two or three over time. You can also compare the answers between them if you ask the same question to each of them to find out which one you prefer.

Specific use cases that work

Drafting emails, particularly the difficult ones. "I need to tell a long-standing client we're putting our prices up by 12%. Help me draft an email that explains the rationale, acknowledges the impact, and doesn't sound apologetic or robotic."

Summarising long documents. Paste in a 30 page contract, a long report, or a transcribed meeting and ask for the key points. Good prompt: "Summarise this document in five bullet points covering the most important commitments, deadlines, and any clauses I should flag with a lawyer."

Job descriptions. Far better than starting from a template. "Help me write a job description for a customer support manager at a 12 person SaaS company. Mention that we're remote first, offer good benefits, and value initiative over hierarchy. Include responsibilities, requirements, and a culture section."

Drafting client proposals. Feed it your last three proposals and ask it to draft a new one for a similar project. It'll match your style.

Working through decisions. Genuinely one of the most underused use cases. "I'm trying to decide whether to take on a £40k project that would book out my team for six weeks and prevent us from pursuing two larger leads. Walk me through the considerations I should weigh."

Editing your writing. Paste in something you've drafted and ask it to make it shorter, clearer, more direct, less corporate, or more friendly.

Researching a topic quickly. "Give me a five minute briefing on what's happened in [industry topic] in the last six months. Focus on what would matter to a small business operating in this space."

Creating PowerPoint presentations. "Create me a slide deck using this data, summarising and providing the back story and our methodology."

What it costs

Realistically, £20 a month per chat tool. Most small business owners need one. Some end up paying for two as they get more sophisticated. That's it. There is no enterprise contract, no setup cost, and no annual commitment.

If you're paying more than £20 per person per month for a chat AI, you're either being sold an enterprise feature you do not need, or you're using a thin layer wrapped around one of these tools and being charged a markup.

My honest verdict

This is where every small business owner should start, and where most of the value is for most businesses.

The mistake people make is treating chat tools as a search engine. "What's the capital of France?" Yes, technically it can do that, but you've wasted the tool. The value comes from longer, more thoughtful interactions: drafting something, refining it, asking for alternatives, and having it challenge your thinking.

If you only do one thing as a result of reading this article, subscribe to Claude or ChatGPT at about £20 a month and commit to using it for one work task every day for the next two weeks. By the end of those two weeks, you'll understand AI better than 90% of the consultants trying to sell you something.

The best mental model: this is not a tool, it's a colleague. You would not expect a new colleague to be useful on day one without context. You'd brief them, give them background, tell them what you wanted, react to their first draft, and redirect them. Treat the chat tool the same way. The people who get most value from these tools have learned to treat them like a real working relationship rather than a vending machine.

My top tip

The one word of warning I would give is around chat history. It can only remember so much, and the more you ask it to do the less it will remember of the earlier conversations. This is improving over time. So my top tip is to make sure you give your chats enough context to your request, as this really improves the outcome.

Category 2: Apps

What it is, in plain terms

AI features built into software you're already using. Not a separate tool you go to: an AI capability inside the apps you live in every day.

The defining trait is that the AI is where the work already is. You do not switch applications, you do not copy and paste, and you do not open a new tab. The AI shows up inside Outlook, inside Notion, inside your CRM, and helps you with the task at hand.

The main places you'll encounter this

Microsoft 365 Copilot inside Word, Excel, Outlook, Teams, and PowerPoint. It summarises meetings, drafts emails, builds spreadsheet formulas, and creates first draft slides.

Google Workspace Gemini inside Gmail, Docs, and Sheets. Similar functions to Microsoft's offering, but inside the Google ecosystem.

Notion AI inside Notion. Probably the best implementation of in-app AI right now, particularly for content work, summarising notes, and drafting documents.

CRM systems such as HubSpot, Salesforce, and Pipedrive. They help with follow-up emails, call notes, and reports.

Accounting software such as Xero, QuickBooks, and FreeAgent. These tools increasingly use AI for categorising transactions, generating insights, and drafting customer communications.

Project management tools such as Asana, ClickUp, and Monday. They summarise project status, generate reports, and draft task descriptions.

The pattern is simple: if you use a piece of business software, it almost certainly has AI features now, either available already or arriving in the next six months.

What it costs

This is where it gets murky. Some AI features are included in your existing subscription. Microsoft 365 Copilot, for example, started as an expensive add on but is increasingly bundled. Notion AI is included in higher tier Notion plans. Some CRMs charge a premium tier for AI features.

You can pay anywhere from £0, because it's already included, to around £30 per user per month for premium add ons.

The rule: check what's already included in your existing software subscriptions before paying extra. Most small businesses are already paying for AI features they're not using.

There is a growing trend towards per-token usage, which basically means the text you type is charged for in batches. Every vendor is different, but we will see more token-usage-based billing over the next few years.

Specific use cases that work

Microsoft 365 Copilot in Outlook: summarising a long email thread you've just stepped into and drafting a reply that captures the key points.

Notion AI: drafting documents from rough notes, summarising long pages, and generating action items from meeting notes.

CRM AI features: drafting follow-up emails to leads based on the conversation history and summarising notes from a long sales call into a structured format.

Excel or Sheets AI: generating formulas from descriptions in plain English. "Calculate the percentage change between this month and last month, highlighted in green if positive and red if negative."

My honest verdict

Useful but uneven. Some app AI is excellent and worth paying for. Some is weak and being marketed harder than it deserves.

The genuinely good ones right now include Microsoft Copilot inside Outlook and Teams, Notion AI, the AI features inside Cursor for developers, and Gemini inside Google Docs for collaborative writing.

The overhyped ones include most CRM "AI" features that are really just prompt templates, most accounting software AI that is really just rules-based categorisation, and most project management AI that is really just keyword summarisation.

The rule for this category: trial before you buy. Most of these are sold on the marketing brochure and disappoint in practice. Use the free trial, do real work, then decide if it saves you enough time to justify the price.

For most small businesses, use the AI features that come included in what you already pay for. Do not pay extra for premium AI add ons until you've personally verified they save you real time. The chat tools in category 1 will give you most of the value for less money.

Category 3: Agents

This is the loaded category. I'm going to be especially direct here because the noise around this is causing real damage to small businesses making bad decisions based on marketing.

What it is, in plain terms

AI that can take actions on your behalf, not just give you answers. You give it a task, it goes off and does it: browsing the web, sending emails, working across multiple systems, making decisions, and completing multi step workflows without you having to babysit every step.

This is the version of AI that genuinely could replace some kinds of human work in time. It's also the version that lives in press releases and YouTube videos far more than it lives in the real world right now.

The honest state of play

The technology is real and improving fast. The marketing has run ahead of the technology by about a year.

The agent products that work reliably right now are narrow and expensive, usually built for specific use cases in specific industries, sold to enterprise customers, and costing thousands per month. They work because they've been hand engineered for a tight problem space.

The agent products being sold to small businesses are usually one of three things:

1. Dressed up automation: a workflow tool with an AI label on it. Useful, but not really an agent. It should cost what automation costs, not what an "agent" costs.
2. A chat tool with extra steps: useful, but the value still sits in the chat interaction rather than autonomous action.
3. A genuine attempt at an agent that does not yet work reliably enough to deploy in a real business. It fails too often, requires babysitting, and ends up costing more in monitoring time than it saves.

The signal to look for

When someone tries to sell you an "AI agent", ask them this: "Can you show me a small business that's been running this agent successfully for at least six months, without significant human intervention?"

If they cannot, they're selling you a beta product. Be your own beta tester if you want, but pay accordingly.

If they can, take it seriously. The narrow agent products that genuinely work for specific use cases are starting to emerge, particularly in customer service and basic sales operations. But they are still rare.

My honest verdict

For most small businesses today: do not spend money on this category yet.

I'm not saying ignore it. Read about it. Understand who the serious players are. Pay attention to which agent products are emerging from real engineering versus which are emerging from marketing departments.

Do not be the small business owner who pays £500 a month for an "AI sales agent" that fails 60% of the time and damages your customer relationships. That happens far more than people admit.

The real agent moment is coming. When it arrives, probably within 18 to 36 months, it will be transformative. Until then, the smart move for a small business is to develop strong AI literacy in category 1, chat, and category 4, coding, so that when the agent moment actually arrives, you've got the foundation to take advantage of it properly.

A lot of the larger chat tools now integrate an element of agentic running. In other words, they use the frameworks of agents running behind the scenes to help with autonomous running. These agents are safe to use and should be encouraged where practical.

Category 4: Coding / building

What it is, in plain terms

Using AI to build actual software for you: websites, apps, internal tools, and automations, without hiring a developer.

The trendy name is vibe coding. The grown up name is AI assisted development. They mean the same thing. You describe what you want in plain English, the AI writes the code, and you get a fully working something at the end.

The main tools worth knowing

• Lovable: building web apps and sites from scratch, no coding required
• v0 by Vercel: building UI components and prototypes
• Replit: full apps, with hosting included
• Cursor: best if you have some coding experience and want to go deeper
• Claude Code: command line workflows for the technically inclined

For a non technical small business owner, Lovable or v0 are the right starting points. They're designed for people who do not code.

Specific things you could build

A custom booking system for your service, integrating with your calendar and sending automated confirmations.

An internal dashboard that pulls key business metrics into one screen: sales pipeline, cash position, project status, designed exactly how you want it.

A customer portal where clients can see the status of their work, submit requests, and access documents.

A lead capture form that's smarter than what your website builder offers, with qualifying questions, automatic routing, and CRM integration.

A simple inventory tracker built around how your business actually counts stock, not how generic inventory software thinks you should.

A team training site for your standard operating procedures, with progress tracking.

Any of these would have cost £5,000 to £20,000 to commission from a developer two years ago. You can now build them in an afternoon for the price of a £20 subscription.

What it costs

Most of these tools are around £20 to £30 per month. You typically need one to get going. The actual usage costs, the AI tokens needed to generate code, are usually included in the subscription up to a generous limit, then charged usage based beyond that.

For most small business use cases, about £20 to £50 per month covers everything you'd actually do. The expensive end is for serious commercial software development, not for a small business building internal tools.

My honest verdict

This is the most underrated category for small businesses right now, and where I'd urge you to put the most curiosity.

The reason is that this is where small businesses get a structural advantage over big ones. A big enterprise cannot just have someone vibe code an internal tool in an afternoon, because they have procurement, security reviews, and IT approvals. A small business owner can sit down at lunchtime, describe the tool they need, and have it working by dinner.

That's a real shift. Small businesses have always been more nimble than big ones, but they've usually had less technology. AI coding tools flip that. You can now have more technology, customised more specifically to your business, faster and cheaper than many enterprises can manage.

The catch is that you do need to develop some basic understanding of how software works to use these tools well. Not coding skills, just an instinct for what's reasonable to ask, how to describe what you want, and how to iterate when something doesn't work. That instinct comes from practice. Build three small things and you'll have it.

The starting move: pick one small annoyance in your business. A spreadsheet that's painful to maintain, a process that requires too much manual data entry, or a report you wish existed but does not. Try to build a tool for it using Lovable or v0. Budget a Saturday afternoon. You'll be surprised how far you get.

How to use this framework

Once you've got these four categories in your head, here's how to actually use them.

The question to ask every AI vendor is: "Which category is this?"

Is it chat? Is it app integration? Is it an agent? Is it a coding tool? If they cannot give you a clear, single category answer, that itself is information. Most genuinely useful tools fit cleanly into one category. The products that try to be everything to everyone are usually the ones that do not work properly at any of it.

The pricing rule

1. Chat: about £15 to £25 per month per user
2. App integration: £0 to £30 per month, often included in existing software
3. Agent: hundreds to thousands per month, and only if proven
4. Coding: about £20 to £50 per month per user

If someone's pricing does not match the category they're selling, that's a red flag.

The attention rule

If you've got an hour a week to invest in AI:

1. Spend it on category 1 first until you're fluent, roughly four to eight weeks
2. Then explore category 2 as you encounter AI features in your existing software
3. Get curious about category 4 once you've got some confidence
4. Keep category 3 as background reading: interesting to watch, dangerous to invest in until it's more mature

Where to start

If you're reading this and wondering where to actually begin:

First two weeks

• Subscribe to Claude or ChatGPT
• Pick three real work tasks from your current to do list
• Try doing each of them with the chat tool, paying attention to what works and what doesn't
• Notice which app features you already pay for that have AI built in

Weeks three and four

• Pick one small annoying thing in your business
• Try to build a tool to fix it using Lovable or v0
• Do not worry if it does not work. The attempt itself teaches you more than reading about it.

This quarter

• Develop genuine fluency with one chat tool
• Audit which AI features you're already paying for in your software stack
• Build one or two small internal tools
• Stay informed about agents without spending money on them

That's the playbook. Four moves, four categories, twelve weeks.

The bigger picture

The reason terminology matters this much is that vocabulary is the gateway to everything else.

Until you can sort the AI conversation into categories, you cannot make good decisions about what to use, what to pay for, what to ignore, and what to invest your scarce attention in. You're at the mercy of whoever's selling you the loudest thing.

Once you can sort it, you become the person in your business, and in your network of fellow founders, who actually understands what's going on. That's a real asset. The small business owners who develop genuine AI literacy in the next twelve months are going to be operating with a meaningful advantage over those who do not. Not because they'll know more about AI as a technology, but because they'll be able to think more clearly about how to use it.

That's what this comes down to. Not hype. Not doom. Just clarity.

If this guide was useful, the best way to support the show is to share it with one small business owner who needs it.