10 AI Tools That Are Changing the Way Creators Work

There are now somewhere north of 14,000 AI tools marketed at creators. The honest truth is that we’ve permanently adopted around ten of them. The rest fall into one of three categories:

1. Wrappers around an LLM with no defensible workflow.
2. Genuinely good tools that solve a problem we don’t have.
3. Tools we’ll use once a quarter and forget exist.

What follows is the small set that has actually changed how we work day-to-day.

Disclosure. Some of the tools below have affiliate programs. We have not enrolled in any. Where pricing is mentioned, it’s the rate we’re paying.

1. Claude (Anthropic) — for editing, not first drafts

We use Claude every day, but almost never to write from scratch. The strongest use is “edit this draft to be 30% shorter without losing the argument”. Pasted into a 3,000-word essay, it consistently produces a tighter version that requires only minor passes.

Why it earns the subscription. Long-context handling is the best in the field. We can paste an entire research dossier and ask for synthesis without chunking.

2. Descript — for podcast and video editing

Edit video by editing the transcript. Once you’ve worked this way, going back to a timeline editor feels like quill-and-ink.

Why it earns the subscription. Removes filler words and dead air automatically. The Studio Sound feature alone is worth it for anyone recording in untreated rooms.

Watch out for. The auto-captions are good but not flawless. Always run a human pass before publishing.

3. ElevenLabs — for voice work

For dubbing, voice-clones, and audiograms. The latest multilingual models are convincing enough that we no longer book a Spanish or German voice actor for short-form content.

Why it earns the subscription. Consistency. You can re-render a line without re-recording.

Ethics note. We only clone voices we have written permission to clone. We won’t cover tools that don’t enforce the same.

4. Runway — for B-roll and motion graphics

For 90% of B-roll, generative video has crossed the line from “novelty” to “cheaper than a stock-footage subscription”.

Why it earns the subscription. The new motion-brush features mean you can animate a still photo (or a generated image) in ways that would have taken hours in After Effects.

5. Topaz Video AI — for upscales and restoration

Not generative, but AI-powered. We use it to bring older interview footage up to publishable quality and to deinterlace tape-era source material.

Why it earns the subscription. No browser-based competitor is in the same league. Worth the one-time license.

6. Granola — for meeting notes that don’t suck

Granola sits silently on your laptop, transcribes your meetings, and writes a structured summary against the rough notes you typed during the call.

Why it earns the subscription. It’s the rare AI tool that produces output more useful than what you typed yourself, because it’s grounded in your notes rather than just “summarize this transcript”.

7. Perplexity — for first-pass research

We’ve almost stopped using Google for research-heavy questions. Perplexity returns sourced summaries, fast, with the option to drill into the citations.

Why it earns the subscription. Pro Search and the focus modes (“Academic”, “Reddit”) genuinely change what’s possible to find.

Watch out for. Always click through to verify a citation. Sourced doesn’t mean correct.

8. Cursor — for the technical creators

If part of your work is shipping software (a microsite, a Next.js project, an automation), Cursor is the editor we’ve replaced VS Code with. AI-assisted coding that feels like pair-programming, not autocomplete.

Why it earns the subscription. Multi-file edits are the killer feature. Asking the editor to refactor across the project actually works.

Related ecosystem. If you ship a modern web app or marketing site (for example, Next.js, lead magnets, in-browser tools like ours), the open skills.sh directory is worth bookmarking. It lists agent skills—reusable procedure packs you install with a single command—not a Prelink product, and we don’t operate the registry. A fuller curated list is in the section below so you can match skills to the kind of work we describe on this site.

9. Topology — for thumbnail iteration

Generate a hundred thumbnail variants in the time it takes to make one in Photoshop. We don’t ship them straight, but they’re excellent inspiration starters.

Why it earns the subscription. Composition variations on a single concept. We routinely take a generated thumbnail and rebuild it by hand at higher fidelity.

10. Numerous.ai — for spreadsheet brain

A Google Sheets add-in that lets you call an LLM in a formula. We use it to classify thousands of survey responses, summarize transcripts in bulk, and tag URLs.

Why it earns the subscription. Replaces what would otherwise be ad-hoc Python scripts and lets non-technical teammates run the same workflows.

skills.sh: curated picks for a Next.js site + marketing stack

This article is for creators, but a growing share of readers also ship software: landing pages, newsletters, in-browser tools, and lightweight automations. If that is you, the rest of this section is a long-form addendum on skills.sh—an open directory of agent skills you can install next to a coding assistant. It is not a Prelink product: we do not run the registry, and nothing here is sponsored.

What “agent skills” mean in plain English

A skill (in this ecosystem) is not a new SaaS dashboard. It is closer to a playbook in a file: checklists, conventions, and copy-pasteable procedures that a model can load when the task matches. You keep them under version control; you share them the same way you share ESLint config or a design system. The difference from a random prompt in a Notion page is structure and re-use—one source of truth your agent sees every session, instead of rediscovering your house style from a five-paragraph preface you typed in January.

That matters when you are doing the same work categories every week: tightening React components, auditing a route for Core Web Vitals, formatting a lead-magnet PDF, or making sure a new insights post did not break metadata. A skill does not replace the official Next.js or React docs; it sits on top of them as a project-specific how we do it here layer. Think of it as a senior engineer’s notes, encoded so an assistant can follow them consistently.

The install path (at a glance)

Most packages in the directory are meant to be discoverable and installable without a private beta. The exact command varies by how the author ships the package, so treat the skills.sh page for each skill as canonical. A pattern you will see in tutorials is:

npx skillsadd <skill-name-or-slug>

Before you run anything: open the skill’s own README or SKILL.md, check when it was last updated, and confirm the name you see in the leaderboard or search box. The ecosystem moves fast; a slug that worked in a blog screenshot may have been renamed. If your stack is a fork or monorepo, one skill might assume plain app/ routes while another assumes src/app/—that is a normal merge conflict for human judgment, not something to automate blindly.

A one-week tryout that actually sticks

How to use them in practice is the same discipline we use for new AI software elsewhere in this list: pick two to four skills that match a current bottleneck—not every category at once. A realistic week for a solo builder shipping a site like ours might look like this:

• Day 1–2: one UI-oriented skill (for example shadcn or frontend-design) and one vercel-react-best-practices pass on the noisiest components in your repo. Goal: smaller diffs, fewer re-renders, consistent spacing tokens.
• Day 3–4: add seo-audit (or schema-markup) and run it against a single high-traffic template—your home page or a pillar insights article—before you touch the rest. Goal: one honest punch list, not a hundred shallow warnings.
• Day 5: either pdf (if you ship downloadable assets) or programmatic-seo / content-strategy if you are scaling MDX. Goal: a repeatable check before you add ten more near-duplicate pages.

At the end of the week, turn off anything you did not open twice. A skill that never leaves your node_modules of the mind is just clutter.

How this pairs with a “Prelink-style” stack (honestly)

Our site is a Next.js editorial app with MDX content, a tools hub, and the occasional PDF or static asset. We do not wire skills.sh into our production build; readers never execute a skill in the browser when they use our calculators. The connection is for creators who also code: the same people who outgrow ad-hoc ChatGPT threads often outgrow ad-hoc shell scripts and unmarked TODOs in README files. Skills are a middle ground—more durable than a prompt, lighter than a microservice.

If you maintain newsletters and sponsorship math, pairing seo-geo or ai-seo skills with our engagement rate and sponsorship rate calculators is a sane split: the skill keeps language and structure consistent; the tool keeps the numbers defensible in a deck.

Trust, scope, and what to skip

Skills are community content. Treat them like any other dependency: skim the source, do not run unknown post-install scripts on a machine with production secrets, and use a branch. If a skill nags you to add an MCP server or a cloud key you do not need, ignore it—mcp-builder and similar entries are for people building integrations, not for shipping a read-only marketing site.

How to use them (short version): pick a few that match this quarter’s pain, run the documented npx skillsadd (or equivalent) for the exact name you verified on the site, use them for a week, then add or drop. The leaderboard on skills.sh remains the source of truth for install counts and naming.

Building the product (Next / React / UI)

• vercel-react-best-practices (or similar “React best practices” from Vercel) — code quality, patterns, and common footguns.
• shadcn / shadcn-ui — if you use shadcn, this keeps components and theming consistent.
• frontend-design / web-design-guidelines — layout, typography, and accessibility-minded UI polish.
• next-cache-components or other Next-related skills — App Router, caching, and newer Next.js patterns when you are on the bleeding edge.

SEO, content, and being findable

• seo-audit — a structured pass over technical and on-page SEO.
• programmatic-seo / content-strategy — if you are adding many pages or content at scale.
• schema-markup (if listed) — JSON-LD and structured data.
• ai-seo or seo-geo — when you care about both classic search and AI-surface visibility.

Files and assets (for example, lead magnets)

• pdf — handy when you generate or review PDFs, similar to how we script PDF output for downloads in our own stack.

Meta: finding and writing skills

• find-skills — helps discover or compare skills in the ecosystem.
• skill-creator / writing-skills — if you want to author your own skills well.

Workflow and quality

• verification-style or test-driven skills — if you want stricter “check before you ship” loops.
• mcp-builder — only if you are building MCP servers or tools, not for a plain marketing site.

Prelink and skills.sh: we point readers to the directory because the same people who outgrow ad-hoc prompts often outgrow ad-hoc, one-off procedures in code. Versioned, reusable skills are a useful mental model—whether or not you adopt every package on the list.

What we tried and dropped

A short rogues’ gallery, in case you’re considering them:

• Notion AI — useful, but underwhelming compared to opening Claude in another tab.
• Jasper — once category-defining, now hard to justify next to direct LLM use.
• GenAI thumbnail tools that promise “100 viral thumbnails per click” — uniformly bad without manual rework.
• AI “clip extractor” tools for podcasts — the auto-selected clips were consistently the wrong moments. We pick the clips ourselves.

How to evaluate a new AI tool

A short test we now run before subscribing:

1. Is the output closer to a draft or a deliverable? Tools that produce rough drafts only earn a place if the “rough” is meaningfully faster than starting from scratch.
2. Is it grounded in your data, or is it generic? Tools grounded in your data (transcripts, sheets, code) compound. Generic tools commoditize.
3. Do you trust the company in 18 months? AI tooling is consolidating fast. We’re wary of betting workflow on tools likely to be acquired and shuttered.

The ten above pass all three for us. We’ll re-run this list at the end of the year.