20 AI Tools From One Curator: A Practitioner's Review of the Applied AI List

Curated AI tool lists are everywhere, and most of them read like warehouse catalogs: fifty logos, zero opinions. The Applied AI list by Matt Penny is the opposite kind: twenty tools, short honest descriptions, no ratings theater. That brevity is exactly why it is worth a practitioner’s review, because with a list this tight every slot was chosen.

We run an automation studio, so our test for any tool is simple: does it already do billable work, or is it a demo? Here is the Applied AI list against that standard.

Four that already run our production

Remotion. An open-source React framework that renders videos from code. This is not a “someday” tool for us: our article-to-video pipeline renders 30-second vertical reels from article data, with the palette, fonts and timing living in components. When the brand changed, the fix was a token edit and a re-render, not a timeline re-edit. If you write React and need recurring video, nothing else on this list comes close.

Nano Banana 2. Google’s fast image model, and the engine behind our blog cover production: sixteen-plus covers generated through a wrapper service, each checked and cropped to a strict 1200x630 frame. It is fast enough that cover generation stopped being a milestone and became a step. The honest caveat: text rendering in images is still its weak spot, so we keep lettering out of prompts.

Hyperframes. The other “video as code” framework on the list, HTML/CSS/GSAP based instead of React, open source under Apache-2.0 with local rendering. We ran it for a promo clip, and it is a serious option when your team’s native tongue is the web stack rather than React. Two video-as-code tools on one curated list tells you where the curator thinks media production is going: deterministic, scripted, reviewable.

Terminal coding agents (the Claude Code class). Our studio literally runs on one: this article, the deployment pipeline, the social publishing stack, all written, reviewed and shipped by an agent in the terminal with a human approving the deltas. Claude Code is the best-known name of the class; OpenCode and Roo Code are the open-source flavors, and the differences today matter less than the workflow shift: agents write, humans decide.

On our bench, seriously evaluated

n8n. Workflow automation with 500+ integrations and a visual editor. Solid for teams whose automation owners are not developers. Our conveyor is code-first and event-driven, so n8n is the alternative we recommend to clients, not the one we run. Both are legitimate answers; know which kind of team you are.

OpenRouter. One API, 400+ models. Attractive as a hedge and for sampling new models. We stay on direct provider APIs for production because billing, rate limits and data paths stay explicit. For experiments and benchmarks, a router is hard to beat.

Jina.ai and Tavily. Search infrastructure for RAG and agents: embeddings, rerankers, readers, agent-oriented search APIs. If your AI plans include “chat with our documents”, these two names belong on your shortlist before any chatbot vendor.

Antigravity, Lindy, Airtop. The agent-platform wave: an agent-first IDE from Google, no-code AI employees, no-code web agents. We watch this space with engineer’s skepticism: the moment an “employee” touches real systems, auditability beats demo polish, and code-first agents currently win on that axis. But the direction is right, and for teams without engineers these are the most accessible doors.

The rest, at a glance

GenAI Pro bundles text, image and voice under one subscription as an alternative to several premium plans. Higgsfield and Google Flow cover AI video, the former with realistic human characters, the latter pairing Nano Banana 2 images with Veo 3 editing. Wispr Flow speeds up writing by voice. CapCut is the free short-form video editor everyone already knows. OpenArt is a friendly image-generation front end (its link on the list carries affiliate parameters, worth knowing). OpenClaw runs an autonomous agent locally, wired into messaging.

None of these are bad tools. They simply are not our bottlenecks, and that is the whole point of the review method.

The verdict

The Applied AI list earns its keep: short, current, and unsurprisingly heavy on agent tooling and video-as-code, which matches where production work is actually moving. Our four-in-production score is unusually high for a curated list, which says the curation is grounded in real use, not press releases.

The method matters more than the list. Filter by bottleneck, prefer tools that compound (code, templates, pipelines you keep), and run a two-week pilot on a real task before giving anything a permanent seat. A tool list is a menu. Your production stack is a diet, and diets are personal.

Original list: its-applied-ai.com/tools. We build the kind of pipelines these tools plug into, agents, media and integrations first: automata.sale.

Phone: +7 (906) 311-77-69, email: hello@automata.sale, Telegram: @automatasale, site: automata.sale

Evgeny Uryadov, sole proprietor (INN 645112058391, OGRNIP 312645301900058). Working across Russia, Belarus and Kazakhstan.