The playbooks.

There are 200-plus AI agent frameworks and the listicles rank them like a leaderboard. That's the wrong lens. Underneath the branding they come in three shapes — a loop, a graph, or a crew — and the shape is what you're actually choosing. Here's each framework builders reach for in 2026, sorted by shape, with the exact command to start and the honest catch for each.

Every AI code review tool promises the same thing — catch more bugs, merge faster. The claim is half-true, and the marketing hides the one difference that actually matters for vibe-coded apps: does the tool read your whole codebase, or only the changed lines? Here's what CodeRabbit, Greptile, Qodo, Codex, and Ellipsis actually do, the only numbers worth trusting, and when a solo builder should skip the subscription.

There's no single best AI agent — four in-depth 2026 round-ups picked four different winners, because "best" depends on the job. This guide sorts the field by what you actually need: an always-on assistant, a hands-off research agent, a browser task-runner, workflow automation, or a framework to build your own — with the honest pick for a one-person business and where each one bites.

Every major AI coding tool now speaks MCP — Claude Code, Cursor, VS Code, and, since 2025, OpenAI's and Google's tools — so the real question isn't whether to use MCP servers but which, and how few. This is a field guide organized by the job you're doing, from your database to browser testing, with the honest pick for each and why a small, curated set beats installing everything.

If you have ADHD and a folder full of half-built projects, the problem was never discipline. AI coding agents change the equation: they absorb the parts of building that stall an ADHD brain — starting the task, writing the boilerplate, holding the whole codebase in your head — and leave the parts it's good at, hyperfocus and fast, novel ideas. This is what's actually going on, and how to build with it instead of against it.

Claude Code can run a real marketing loop for the app it built: connected to the Google Ads API it creates campaigns, writes ads, and audits search terms; pointed at your own repo it ships one landing page per search intent and fixes whatever Lighthouse flags. What it can't do is know search volumes or spend your budget wisely — that part stays human.

Claude Code is an AI agent, not a chatbot: it doesn't just answer questions, it edits files, runs commands, and ships working software from plain-English instructions. This guide explains every concept you'll meet in your first month — context windows, tokens, CLAUDE.md, skills, MCP, permission modes, checkpoints — in plain words, one analogy each.

Nearly every recurring SaaS charge in a builder's stack has a production-grade open-source twin: Cal.com for Calendly, Documenso for DocuSign, n8n for Zapier, PostHog for per-event analytics, Coolify for platform bills. The honest rule before you migrate anything: self-hosting kills the license fee, not the work — you become the admin.

The way to make money with AI in 2026 isn't building clever agents for yourself — it's selling automation to businesses that need it. The technical build is commoditizing fast (Claude Code and n8n turn months of work into days), which means the scarce, valuable skill is finding clients, understanding their problems, and pricing on the value you deliver — not the time it took.

Claude Opus 4.8 leads the benchmark aggregate (about 86 to 80) and wins the coding and hard-reasoning tests; Google's Gemini 3 Pro answers with a larger 2M-token context, native multimodality, and roughly half the token price. Pick Opus for the hardest reasoning and real-repo coding; pick Gemini for cost, huge context, and grounded multimodal work.

Cursor and Windsurf now cost the same (~$20/mo) and both are strong AI-native editors. Cursor is a polished VS Code fork that wins on complex, single-developer work; Windsurf runs inside 40+ IDEs, has the autonomous Cascade agent, its own fast SWE-1.5 model, and deeper enterprise compliance. Pick Cursor for depth in one editor; pick Windsurf for mixed-IDE teams and compliance.

Claude Code is a terminal-first autonomous agent that runs Anthropic's own models with a 1M-token context and orchestrates teams of agents; Cursor is an IDE — a VS Code fork — with the best tab-completion and its pick of frontier models. The practical split: Claude Code for long, autonomous, multi-file work (and it's far more token-efficient), Cursor for fast, interactive, in-editor coding. Many developers run both.

GPT-5.6 is OpenAI's new three-tier model family — Sol (flagship), Terra (balanced), and Luna (fast and cheap), all with a ~1M-token context. Sol tops the agentic-coding and terminal benchmarks and costs about half of Claude Fable 5, but Fable 5 still leads on SWE-bench Pro — the test closest to real GitHub work. The honest split: reach for Sol in an agent harness and terminal, and for Claude on a big real-world repo.

AI writes worse code when you give it more context, not better — a bloated context confuses the model, burns tokens on every request, and sends it editing the wrong files. The fix is spec-driven, context-efficient coding: write a short spec (exact stack versions, architecture, the reusable code to use, and how to verify), load only the documents a task needs, and work in small sprints with a fresh context each time. AI-driven development is really context management.

A faceless YouTube channel shows no face and no camera — just a voiceover over AI-generated images, stock B-roll, or animation. In 2026 the whole pipeline runs on AI: pick a validated niche, write the script with Claude or ChatGPT, narrate it (your own voice is safest), generate one visual per beat, and auto-edit the cut. The winners aren't the ones with the best tool — they pick a niche that has buyers before they publish a single video. It works the same on Reels and TikTok.

You do not need a credit card to build with LLMs. Google AI Studio has the most usable free tier, Groq and Cerebras are the fastest, OpenRouter gives you one key to dozens of free models, and Mistral hands out up to a billion tokens a month. This guide pairs each free API with five concrete jobs it's best for — so you pick by the work you're doing, not the hype. Free tiers are for prototyping, not production; treat them accordingly.

A good CLAUDE.md file is short, specific, and ruthlessly curated — because it loads into Claude Code's context on every prompt, and past a couple hundred lines it makes Claude worse, not better. Put the project's stack, key commands, conventions, and hard "do nots" in it; move everything else to imports, rules, or skills. AGENTS.md is the same idea as an open standard that works across Codex, Cursor, and 20+ other tools — and Claude Code can read it too, through a single import line.

The best Claude Code prompts are specific, hand it real context, and plan before they touch code. This is a curated copy-paste library organized by how you actually work — understand a codebase, plan, build, review, ship, and operate — built on Anthropic's official prompt library, with our own examples and notes on when to reach for each. Plus the slash commands (/init, /plan, /compact, and custom ones) that make a session faster.

Product Hunt is still the biggest launch day, but its links are nofollow and the front page is saturated — so most makers pair it with alternatives. The best in 2026: Uneed and Fazier for an engaged indie and AI audience, Startup Fame for a confirmed dofollow link, MicroLaunch and Peerlist for a week- or month-long window instead of a one-day sprint, and Show HN or the right subreddit for raw traffic. Match the platform to your product and your audience's timezone.

The best AI API depends on what you're building. For text and agents, Google's Gemini is our top all-round pick, with Moonshot's Kimi, Z.ai's GLM, and MiniMax's M3 as cheap near-frontier options — and OpenRouter as one key for all of them. For images, Google's Nano Banana and FLUX lead; for video, Google Veo; for music, ElevenLabs. And for anything visual or audio, most developers ship through a host like fal.ai (or our own AIMLAPI) instead of wiring up each provider.

Our pick for the best AI coding tool in 2026 is Claude Code, with OpenAI's Codex a close second — the two frontier coding agents now going head-to-head. Cursor is the nicest editor to work in, but it bills by metered usage and still leans on the same third-party models under the hood, so heavy use gets pricey next to the two. For non-coders, builders like Lovable and Replit turn a prompt into a running app.

Coding your app is the easy half now. The copy that sells it is the other, and the moves that built decades-long, category-defining campaigns still work: lead with one line that lands a punch, open a curiosity gap, tell a story where the customer is the hero and your tool is the guide, cut every draft by a third, and get good by copying great writing out by hand. Here are eight legendary ads and exactly how to steal each move for your app.

The apps solo builders actually ship fast fall into six repeatable types — a one-feature identifier, an unbundle of a bloated tool, a content utility, a community-branded consumer app, a polish-wins-the-category app, and a low-ticket mini-product. But the code is the cheap part now. The hard part is picking a real Job, keeping the offer to one thing, validating before you build — and then promoting it, because an app nobody can find makes $0.

Neither wins outright. Codex is faster, leaner on tokens, and sharper at code review; Claude Code is stronger at front-end design, deep customization, and its 1M-token context. The move serious builders keep landing on: run both — plan or build with one, then have the other do an independent review.

If you built something with AI, posting is getting hard to ignore. GaryVee's 'interest media' shift is real — more feeds now recommend content by interest, not only by who follows you — so a maker with zero followers can sometimes out-reach an established brand. That's the unlock and the mandate: building got faster, getting seen is the job, and many products struggle to grow without it. Here's the honest first move, written for a builder who would rather ship a feature than post.

Claude Code can do most of your SEO — build a crawlable static site, write and optimize pages, fix the technical layer, even bottle the whole workflow into a reusable skill. What it can't do is invent authority or pick keywords, and if you let it generate 10,000 pages you'll get de-indexed, not ranked. Here's the honest, hands-on version for a maker whose site shipped last week.

When AI makes building a product trivial and floods the web with generic content, marketing inverts: distribution becomes the moat, and distribution increasingly means being the source AI answer engines and buying agents cite and recommend. The durable move before 2027 is to engineer your product to be machine-discoverable and corroborated from day one — even at zero domain authority. Strong in the future means cited, not ranked.

You don't start by building. You start by finding the one assumption that would kill the idea, then build the smallest thing that tests it. Here's the full idea-to-deployed workflow from a live AI build session: research the market with ChatGPT, pick a segment on unit economics, have AI write the PRD, then vibe-code and ship with Claude Code — in a couple of hours, not a quarter.

A harness is the runtime wrapper that turns a bare language model into an agent — the layer that runs tools, holds memory, assembles context, and enforces limits. The model does the reasoning; the harness does everything the model can't do on its own. It's the part that decides whether you shipped a chatbot or a real agent — and it's why the same model feels brilliant in one tool and useless in another.

'Learn AI' is generic advice. The skills that actually compound as AI improves aren't prompts — they're six capabilities you can start this weekend: running agents, building distribution, building hardware, curating with a real point of view, shipping-and-selling as one person, and pulling people into real rooms. Pick one and get dangerous; pick three and you're the person every team wants.

Don't buy a bot that 'closes the whole sale' — it won't, and people hate talking to one. What works: export your own sales chats, strip the personal data, and run them through a Claude Code agent that audits how you actually sell and writes the follow-ups your team forgot to send. Then put an agent only on the hours you're offline.

The best vibe-coding tools aren't another AI model — they're the open-source Claude Code repos you stack on top of it. The right ones turn the agent from a fancy autocomplete into a disciplined senior engineer: it plans before it codes, designs properly, maps your codebase, and gets scanned for security. Here are eight worth installing.

The expensive mistake in the vibe-coding era isn't building the app — it's building the wrong one. You validate an AI app idea by talking to about eight people in your target market before you build, and listening for the single capability they all wish existed. That's your wedge. Higgsfield's founder did exactly that on the way to a ~$200M run-rate.

Going viral isn't luck — it's a format that earns watch time, run at volume. The founder of nomadtable grew it past a million downloads, solo, by ignoring view counts and chasing one number: how many people keep watching past the first three seconds. Here's that UGC playbook, with his real figures attributed throughout.

The cheapest way to get your first users for a vibe-coded app is organic short-video: post native-looking TikToks and Reels from accounts that read like your target user, and convert people in the comments — no ads, no influencers. It's slow at first and most clips flop, but it's how indie consumer apps go from zero to their first thousands of downloads.

Claude Code has built-in memory — a CLAUDE.md file it reads each session — but it only holds short, hand-written notes. To make it understand a large codebase or docs set, you add a second brain: turn the repo into a knowledge graph, fold that into an Obsidian vault, and let the agent query the map instead of re-reading every file.

To get cited by ChatGPT and Perplexity, be the clearest answer in several places at once. Publish answer-first pages with extractable facts, get listed on directories and review sites the engines crawl, and build consistent mentions across Reddit, YouTube, and your own site. Perplexity tends to favour recent content; ChatGPT appears to weight agreement across sources.

Launch on Product Hunt by preparing for a few weeks: build genuine account activity, line up your assets and a supporter list, and ship at 12:01 AM Pacific (the daily reset) on a weekday. Reply to every comment all day, never ask for upvotes (say 'check it out'), and keep working the week after — that's where most of the lasting value is.

Reddit and Hacker News send the most engaged users, but both punish self-promotion. Post where your specific user already is, lead with the problem and a real story, make your product trivially easy to try, and never ask for upvotes or have friends pile in. On Hacker News, use a plain 'Show HN' title with no hype.

Submit your AI tool where people already search for one. Start with a niche AI tool directory (There's An AI For That, Futurepedia, and Vibedonalds for vibe-coded products), add a couple of general startup and SaaS directories, launch on one platform (Product Hunt or a smaller one like Uneed), and post in one community (a relevant subreddit or Show HN). Spread more directory submissions over the following weeks to get your first users.

The closest v0 alternative is Bolt.new — it generates and runs a full app in your browser on any framework, not just v0's Next.js + shadcn/ui. Lovable is the React + Supabase pick, Replit Agent builds and hosts in one workspace, and a0.dev targets native mobile. Every option here has a free tier.

The strongest GitHub Copilot alternative for agentic work is Cursor — an AI-native editor whose multi-file agent goes further than Copilot's. Windsurf is the other agent-first editor. For free unlimited completion, use Codeium; for open-source, bring-your-own-key control, try Cline, Aider, or Continue. Every option here has a free tier.

The closest Cursor alternative is Windsurf — the other AI-native editor, with a multi-step Cascade agent and a more generous free tier. To leave the IDE for a terminal agent, use Claude Code or Aider. For free, open-source, bring-your-own-key tools, try Cline or Continue. Every option here has a free tier.

The closest Lovable alternative is Bolt.new — it runs the same chat-to-app workflow on any framework instead of Lovable's fixed React + Supabase stack. v0 suits teams already on Next.js and Vercel, and Replit Agent builds from a phone. Every option here has a free tier.