😼 Notion's cutest launch

The Roundup

by Product Hunt

Sunday, Jan 12

Happy Sunday!

Hi, hello — welcome back to another edition of The Roundup, our weekly newsletter covering the highlights of last week on Product Hunt. This week, we've got our usual highlights from the leaderboard, plus an excerpt of an internal discussion on our team about AI dev tools and a new article about how (and why) to build tiny viral apps to help your product take off. Let's dive in.

Weekly

Leaderboard highlights

Grok for iOS — The official app by xAlGrok is an Al-powered assistant, developed by xAl, designed to be maximally truthful, useful, and curious. It seems to have way fewer guardrails on it than ChatGPT or Claude.

AnyParser Pro — Parse multi-language images and documents into JSON/markdownAnyParser Pro provides multi-language document and image parsing. You can use it to extract text, tables, or charts from PDFs, docx, PowerPoints, and images. The best part? It lets you process thousands of mixed-format documents at the same time

Ragie Connect — Build RAG applications on your user dataRagie Connect lets you build RAG applications specifically on your user data. It handles auth and automatic user data-syncing from Google Drive, Salesforce, and Notion, among other apps.

Notion Faces — Bring your custom self-portrait in Notion, officiallyNotion has officially launched their own avatar creator. It lets you customize every detail of your avatar — from skin tone to hair cut — with their signature style.

Humiris - Mixture of AI — Custom reasoning model for accurate AI agentsHumiris automatically optimizes the accuracy & cost of your GenAI models. It lets you integrate GPT4o, Sonnet 3.5 and more with your custom models to achieve improved accuracy compared to general reasoning models.

Show more products

Overheard in the community

Was 2024 the year of AI dev tools?

On our Golden Kitties Slack channel, our CEO (Rajiv Ayyangar), CTO (Mike Kerzhner) and staff engineer (Matt Carroll) briefly debated whether 2024 was the year of the AI developer tool. We found their discussion fascinating, and feature an edited excerpt of it below.

Rajiv: Was 2024 the year of AI dev tools? E.g. Supabase, Cursor, Replit…

Mike K: It has been the year of “All knowing, sloppy engineer at your fingertips.”

Rajiv: As a non-developer, I’m envious of how much AI seems to have accelerated coding. I don’t feel AI has changed my life that much. Maybe this coming year it will. cough cough, I’m looking at you Siri.

Matt: More of a meta discussion, but I’ve found it hard to really leverage AI unless I know what is happening and I’m quickly able to audit / fix up.

Mike K: We no longer program. We just give feedback to an eager, fast, and loose intern

Matt: Yes, exactly. I recently learned a new language that was pretty unfamiliar and letting the AI “run free” was rough. It would inherently break something and keep digging itself deeper to resolve the problem. Without understanding myself I couldn’t really fix it either.

The flow I wound up using was: Use AI for help with very narrow scope feature, test and integrate it correctly myself (making sure I understand), iterate.

Basically there is this equation: The greater I understand a language the more slack I can grant the AI to do work, because I’ll be able to quickly fix its mistakes.

Mike K: Yup. 2024 has been the year of “developer autocomplete.” We went from autocompleting words to autocompleting functions, classes, features and codebases. But autocomplete only works if you are a good human editor and curator. And it’s hard to become a good editor unless you are a skilled writer.

That’s why I am really curious to see what happens to programmers in 2025 and beyond. Folks who don’t know how to ride a bike are given F1 cars. But it’s hard to drive an F1 car if you have no idea how things in motion behave.

Your agents stopped answering and started doing

Your agents call tools, write to databases and move money, so a wrong answer is now a wrong action, which is a considerably worse meeting. Arcjet runs policy before every LLM call, tool call, query and API request: prompt injection, sensitive data, bots, rate limits, spend quotas, and your own rules in Rego on top. It reads the OpenTelemetry you already emit, so switching it on is one environment variable, no code changes and nothing to deploy. Each check also reads the earlier steps in the same workflow, so an agent's tenth action gets judged against the first nine and not on its own. JavaScript, Python and Go SDKs all at 1.0, agent framework guards included, free tier to start on.

New articles on the site

How to get your first 10 B2B customers

We’ve published a ton of new articles on our site in the past few months. Be sure to check them out!

The Product Hunt interview with Replit CEO Amjad Masad

How we launch at Supabase

The ultimate guide to willingness-to-pay

A Better Way to Get Your First 10 B2B Customers

See all stories here.