wondering.about.ai/ dev studio

I help busy expertsbuild AI tools andcustom software.

Built with you or for you.

Custom AI apps, internal systems, and workflow automation for consultants, agencies, and small teams.

Book a discovery callSee my work
25+
AI tools built
500+
users served
2,800
readers at Wondering About AI
Illustrated portrait of Karen Spinner

About

Hi, I’m Karen Spinner.

I’ve been shipping custom AI software for nearly two years. Some of it is mine, including CarouselBot (live SaaS with paying users), Future Scan (an ML research platform), and several popular Chrome extensions. Some was developed for clients who needed something specific built fast.

As creator of the Wondering About AI newsletter, I also write about how AI development actually works in practice, including original research studies, monthly build challenges, and the failure modes nobody warns you about. Before all this, I spent more than 15 years leading B2B content for technology companies like Adobe and Confluent, growing one agency practice from $200K to $2M.

Services

Three ways to work together.

01 / Build

Need to build something new?

You have a methodology, framework, or workflow that should be a tool. You've thought about hiring a dev team or learning to code yourself. But you don't have time for either, and you want something that captures your expertise and ships in a matter of weeks. I handle copy, design, and code. For highly specialized projects, I assemble a small team.

Book a 20-minute discovery call

How it works

  1. 01NDA and discovery call
  2. 02Proposal with tech stack recommendation, phases, and milestone-based payment
  3. 03Prototype, V0, V1, V2, go live
  4. 04Optional monthly or quarterly maintenance

02 / Finish

Need to finish a project?

You started something. The code mostly works, but the bug list keeps growing, deployment has stalled, or you're stuck somewhere. You'd fix it yourself if you had two clear weekends. (But you don't.) I review your codebase, identify what's actually blocking you, and propose a fixed-scope engagement to get you to ship.

Book a 20-minute discovery callTry the free bug guide first

How it works

  1. 01NDA and discovery call
  2. 02Code review with written findings
  3. 03Proposal with scope, timeline, and milestone-based payment

03 / Coach

Need help getting started?

You have an idea you want to prototype. You want to use Claude Code, but the command line is intimidating, and you're sick of multi-hour tutorials. You don't want to spend a weekend paging through docs. You want someone to walk you through it on your actual project. I coach builders one-on-one. Single sessions for specific blockers. Phased packages for whole projects.

Book a coaching sessionFree reading: vibe coding best practices

How it works

  1. 01Discovery call
  2. 02Coaching proposal with session count or milestone structure
  3. 03Sessions weekly, biweekly, or on demand

Work

Products I’ve built.

CarouselBot

Live SaaS

LinkedIn carousel and document generator with an MCP server. Turns articles into carousels, social cards, and ebook layouts. Live SaaS with paying users.

“One week ago, this CarouselBot post hit, I got a DM, and in less than a week I had a new global client.”

— Rich Carr, Consultant

Future Scan

Closed beta

ML platform that analyzes 150,000+ arXiv papers to surface emerging AI research trends. Semantic search, theme clustering, and automated weekly reports. Built to support my own writing.

“The research tool I've been missing. Surfaces themes, spots patterns, cuts down manual work.”

— Daria Cupareanu, Founder, AI Blew My Mind

Prompt Collector

One-click prompt capture, tagging, and search. Chrome extension.

Substack Reader

Unified dashboard for 1,200+ subscriptions. Chrome extension.

Newsletter Audit

Health metrics and resonance tracking for newsletters. Chrome extension.

Earlier builds

Stack Digest

Archived

Semantic newsletter discovery across 3,000+ publications, with 165 beta users. Built in Python and Django with Celery for async processing.

Read the write-up

Good Bloggy

Archived

An AI writing studio that held a consistent brand voice across drafts. Python with Celery-backed async pipelines.

Read the write-up

Learn

From the newsletter.

Original research and field notes on how AI development actually works in practice.

01

The 7 deadly sins of vibe coding

Common mistakes I see across 25+ builds, with what to do instead.

Read→
02

I ran over 1,000 API calls to find out how ChatGPT grades essays

1,120 calls across four GPT models. AI graders are consistent but exploitable.

Read→
03

Does Claude get desperate when it's running out of tokens?

250 calls replicating Anthropic's interpretability findings at the behavioral level.

Read→
Read more at wonderingaboutai.substack.com→