On My Radar

The most beginner-friendly automation tool I have used, and the one I would still start with today. The visual builder is learnable without a developer background, and once you have built a few flows you start seeing automation opportunities everywhere. Good enough to stay on indefinitely, or a solid stepping stone before moving to n8n or code.

Open-source alternative to Make. More flexibility, more control, and meaningfully more complex to set up. Worth knowing if data sovereignty is a concern, if a developer on the team is already using it, or if you have outgrown what Make can do. For everyone else: Make does the same job with less friction.

Genuinely better than Google Sheets for relational data and client-facing databases. If you are building anything where rows need to link to other rows, or you need a proper front-end for non-technical users, Airtable is where you graduate to. Sheets is fine until it is not.

The category leader for no-code automation. Broader app library than Make; less visual. Worth knowing even if Make is your primary tool.

The tool that made building The Edit possible without a developer. It is not magic. You still need to think clearly about what you are building and prompt well, but it removes the biggest barrier. The GitHub and Vercel integration means you are building on real infrastructure, not a toy platform.

Does exactly what it says and stays out of the way. The auto-deploy from GitHub is the bit that matters. Every Lovable push goes live automatically without you having to do anything. That is the workflow. It just works.

The quiet middle layer of the build pipeline. Every Lovable change writes to GitHub, Vercel watches GitHub, deployment happens without me touching it. Increasingly where the serious work happens as I move into Claude Code. Version control, branch management, and a proper undo button when something breaks. Not optional infrastructure, even if it feels invisible most of the time.

Properly different to using Claude in chat. It reads your actual codebase and makes changes, not hypothetically but actually executing them. Takes some getting used to, but it is the right tool when Lovable hits a wall and you need surgical intervention rather than a full rebuild.

The database layer behind The Edit's subscriber capture, and the right next step when any project outgrows Google Sheets as a data store. Lovable integrates with it cleanly, which matters if you are building without a backend developer. If your site needs proper user accounts, authentication, or relational data, this is the answer.

Google's free playground for testing Gemini, Imagen, Veo, and Stitch before committing to a paid tool. Generous free tier and the fastest way to see what Google's models can actually do. Worth using when you are scoping a build or evaluating whether Google's stack fits a project. Less useful once you have made the decision.

The Vercel alternative worth knowing about. Both hosts do broadly the same job with different developer experience philosophies. Netlify has a slight edge on edge functions and forms, Vercel on frontend framework integration. If you are starting fresh, Vercel is the safer default; if you are inheriting a Netlify setup, there is no good reason to migrate.

The AI-first code editor that most professional developers recommend first, though the more interesting story now is who owns it. SpaceX agreed a £45bn all-stock acquisition of Cursor's parent company Anysphere in June 2026, the largest venture-backed startup acquisition on record, with the deal expected to close in Q3 2026. What that means for the product day to day is not yet clear, and I would watch the next few months before assuming nothing changes. For now it still reads your codebase, understands context across files, and remains the step up from Claude Code when you are writing code daily rather than making targeted interventions. Worth trying when Claude Code starts feeling limited, not before.

The closest thing to seeing what agentic AI actually means in practice. It browses, writes code, and executes multi-step tasks independently, and it is still impressive when it works. The company is now the story: China blocked Meta's ~£1.6bn acquisition in April 2026 and ordered the deal unwound, some employees had already joined Meta, and the founders are barred from leaving China. The product keeps shipping and remains independent for now. I would not build anything load-bearing on it until the ownership question settles.

Free AI-powered UI generation tool from Google. Describe an interface, get high-fidelity screens back using Gemini 2.5 Pro. Pairs naturally with Lovable: design the screens in Stitch, hand them to Lovable to build. The MCP integration is a bonus if you are working in Claude Code; the core tool is useful without it.

A curated library of production-quality React UI components you can drop straight into a Lovable build. Browse visually on 21st.dev, copy the prompt, paste into Lovable. Polished components without a designer. The MCP version for Claude Code is the advanced path; the copy-paste workflow is what most people will actually use.

The industry-standard tool for working directly with APIs. Testing endpoints, inspecting responses, building integrations without writing code first. Learning Postman is a genuine step up from automating with Make.com to working with APIs directly. The free Postman Student Expert certification at academy.postman.com is about three hours and a useful credential to have.

The Lovable alternative for people who want to build in a browser without touching a local environment. The free Starter plan is a genuine test drive. The catch: Replit's credit model means costs can escalate fast with heavy Agent use, so budget carefully before committing to a paid plan.

The open-source answer to Claude Code, and the one to reach for when cost is the constraint. The free tier runs 1,000 requests a day on a Google sign-in, no card. It streams the terminal state rather than asking permission before each change, so it moves faster but watches you less closely. Keep Claude Code for production work where verification matters, and run Gemini CLI for cheap one-shot passes over a large codebase. Worth knowing both.

The low-code AI app builder for people who don't want to touch code. Drag nodes onto a canvas, wire in Claude or GPT, point it at your documents and it handles the chunking and embedding itself with no manual tuning. That automatic knowledge base is the real differentiator. The self-hosted version is free with no meaningful limits. If n8n is the automation workhorse, Dify is where a non-technical team ships a working agent in an afternoon.

The most-starred project in GitHub history, and the one everyone is asking about. It's a personal AI agent you run on your own machine that answers you wherever you already are: WhatsApp, Slack, Telegram, iMessage, and plugs into Claude, GPT or DeepSeek to actually do things: browse the web, run code, control your smart home. The ambition is the appeal. It is also the catch. It runs with real access to your devices and accounts, and its early prompt injection defences were weak enough to make it 2026's first serious AI agent security story. I am watching this one, not running it, and I would tell you to do the same for now. Understand it, because the architecture is where everything is heading. Just don't wire it into your life yet.

Framer has been the professional site builder for years, and the June 2026 Agents release is the reason it finally earns a row. Agents work on the canvas of a live project: they design pages, write components, manage the CMS, and audit SEO, and you can connect Claude Code through the Framer CLI. Compared with Lovable, you trade full-stack flexibility for polish and publishing infrastructure. If the output is a marketing site rather than an app, this is the stronger starting point.

Raycast's app builder went public on 1 July 2026: describe an app in plain language and it builds a real native Mac app, with Claude Code or Codex doing the work under the hood. The apps live on your machine, work offline, and can use the menu bar, shortcuts, and file access, which is what separates this from a web-app builder. There is a community store, so you can install something close to what you need and tweak it rather than starting from scratch. Raycast's six years of Mac craft show in the output quality. macOS only (Apple Silicon), and the free tier is enough to find out whether it sticks.

The AI image tool I reach for when I need logo work or proper brand-adjacent visuals. Vector generation is the genuine differentiator. Most AI image tools hand you a PNG and leave you to figure out the rest. Style locking across a project is what makes it client-ready.

Still produces the best-looking images when creative quality is the priority. The Discord workflow is clunky but tolerable. If you are making campaign concepts or hero visuals and quality matters more than speed, nothing else quite matches the editorial, art-directed feel it produces.

Google's image model, accessible free via AI Studio or paid via the Gemini app. Produces cleaner, more photographic results than Midjourney for realistic imagery; Midjourney still wins on illustration and stylised work. The Pomelli integration for product photography, in UK beta, is the interesting commercial angle.

The mature, professional option for AI product photography. Clean backgrounds, realistic lighting, strong at scale, used by e-commerce teams for a reason. The texture-preservation question is real for any physical product: the test is not whether the image looks good, it is whether it still looks like your actual product. Usually it does.

Renamed from Grok Aurora because xAI's stack moved on: Grok Imagine is now the umbrella for image and video generation, with the Aurora-2 model as its engine since February 2026. Still genuinely competitive with Midjourney for certain styles, and available free with a standard X account. Quality Mode arrived in May 2026 and closed most of the photorealism gap. Fewer content restrictions than most commercial tools impose, which cuts both ways. Worth knowing about even if you use nothing else from Grok.

Bundles 60-plus image and video models into one interface. Recently released a new foundation model. The standout feature is real-time canvas generation: sketch or type and it renders in under a second. Free plan exists but daily credits are limited and commercial use requires paid. Strong for rapid visual iteration; volume-for-money is weaker than buying individual tools directly.

Black Forest Labs' open-weight image generation model. The default open-source option in 2026, used under the hood by many other tools. Accessible directly via Replicate, fal.ai, or local deployment.

The platform a lot of teams run on, and Notion AI sits on top of it rather than beside it. Worth understanding properly if you are advising anyone on workplace AI adoption, because the answer for Notion-first teams is almost always Notion AI, not a standalone tool. The AI layer has improved meaningfully in the last year.

Ad intelligence tool with one distinctive feature: keyword search. Most ad spy tools let you search by brand; BigSpy lets you search by topic or sector. Useful before briefing paid social work, and genuinely eye-opening for seeing what messaging is landing in a market right now.

A macOS menu bar app that gives Claude persistent memory without any setup work. It observes what passes through your Mac, browser tabs, documents, Slack, calls, and indexes it locally, on-device and private. Through an MCP connector, Claude can query that memory so you stop re-briefing it at the start of every session. Free. The obvious comparison is NotebookLM but the use case is different: NotebookLM is a vault you actively fill; minimi is ambient and automatic. Worth trying if you use Claude daily and find context resets your main source of friction. macOS only.

Google's video generation model. Strong on cinematic realism and camera movement, getting better fast. Accessible via AI Studio without a separate subscription, which makes it the easiest entry point for testing AI video gen seriously. Worth watching as the category gap between leaders narrows.

The video model with proper camera controls. You can direct a shot rather than just describe a scene. That distinction matters when you are trying to produce usable output rather than interesting output. Worth trying when a client needs a concept video and a storyboard exists, but a production budget does not.

The speed answer for cutting long video content into short clips. AI decides what is clip-worthy, you approve or reject. Useful for clients sitting on a backlog of webinar or event recordings who need social content fast. Submagic gives you more control; Opus gives you more output.

The benchmark AI video tool. What most serious creators refer back to when evaluating new options. Image-to-video is its most useful feature: take a clean product shot or illustration and add subtle motion, which is a faster route to social video content than shooting from scratch.

The leading AI avatar video tool. The free plan gives you three watermarked videos a month, enough to evaluate whether avatar video fits your workflow before paying. Directly useful for training content, staff updates, and video comms without a camera.

Type a description and get a complete song back, vocals and all. The free plan gives you roughly ten songs a day for personal use. Useful for social video background tracks and reels where licensed music is expensive or complicated. Commercial rights require a paid plan.

Kuaishou's third-gen video model. Launched February 2026. Native 4K, 15-second clips, AI Director for multi-shot sequencing. Strongest motion quality in the category.

An open-source, AI-native video editor for macOS from a YC team, past the four-week bar with active releases through July. The pitch is that you and your agent edit the same timeline: a built-in MCP server means Claude can generate, trim, and re-cut alongside you. Generation with Seedance, Kling, and Nano Banana happens inside the timeline rather than in a separate tab. The editor is free under GPL-3.0; the generative models run as a subscription. Early software, but the direction is exactly where video tooling is going.

The app builder reviewers keep holding up as the direct Lovable comparison. Floot builds and hosts a complete app, backend, database, and authentication included, from a single conversation, and the all-in credit model means you are not juggling separate hosting and database bills the way you can with Lovable. Reviewers consistently flag stronger error handling and fewer stuck-in-a-loop moments than competitors. The trade-off is a single all-in Pro tier at £19 a month rather than Lovable's cheaper entry point, and less granular control once you need something genuinely custom.

A personal AI cloud computer: your own always-on server with an agent that manages files, runs code, and hosts small apps and sites from a single chat interface. The free tier includes 100GB storage and one hosted project; paid tiers start from £13/mo. The pitch is real ownership: your data and tools live on infrastructure you control rather than scattered across five subscriptions, and it sits in the same territory as Manus and OpenClaw, an AI that runs your digital life rather than just answering questions. It suits someone comfortable enough with the idea of a personal server to appreciate what that buys them; it is not yet a one-click product for a non-technical reader. Worth trying if you want a genuinely personal AI infrastructure layer rather than another chat window.