Ari Wilson
Ari Wilson is an individual software engineer, engineering leader, and solo developer/creator rather than an incorporated company. He builds software and engineering organizations, and currently develops Bakeoff, a private evaluation of coding agents on real software tasks, while publishing technical writing and open-source side projects.
At a Glance
- Companies selecting coding tools and agentic engineering workflows
- Engineering teams with proprietary codebases and test suites
- Individual developers and founders who have unused Claude or Codex subscription quota
- Executives seeking network-based job opportunities
- +1 more
AI Tools by Ari Wilson
(1)
RemainderBot
Unused Quota Agent Task Runner
Discussions
No discussions yet
Be the first to start a discussion about Ari Wilson
Latest News
RemainderBot: My part-time assistant runs on tokens I was going to waste — Wilson open-sourced RemainderBot after reporting 111 pull requests over 25 days, with a third merged, a third closed, and a third awaiting review.
Spare a thought for the mathematicians — Wilson wrote about recent AI progress on mathematical proof work.
AI gets great at anything ... if it can check the results — Wilson described rebuilding his WarmPath app and argued that verifiable outcomes enable broader AI capability.
The same bug fix costs 0.4¢ or $2, depending on which coding agent you ask — Wilson published Bakeoff's first detailed results and comparison of model/harness costs, speed, and success.
Products & Services
A private evaluation that takes real bug-fix or feature commits from Wilson's own repositories, hides the tests proving the change, gives the broken tree and prompt to coding agents, and grades the returned work against hidden tests. It is intended to provide outcome-based data for selecting coding agents rather than relying on public benchmarks or intuition.
An MIT-licensed Python package that watches Claude Code and Codex subscription quota, wakes when a weekly reset is near and quota remains, selects a task from configured sources, builds it, and opens a pull request in a private repository for human review.
A personal budgeting app built after Mint shut down. It pulls transactions through Plaid, auto-categorizes them, and provides a personal budgeting agent.
A browser-encrypted password manager whose server never sees the password; originally started on App Engine and later rebuilt.
Market Position
Bakeoff differentiates from public coding benchmarks and vibe-based tool selection by evaluating agents on real, private-codebase bugs and features that models have not seen, with hidden tests and comparable harness measurements. RemainderBot differentiates from conventional autonomous coding services by operating on subscription quota the user already owns, avoiding API overage charges, running in a user-controlled private repository, and requiring human review through pull requests rather than silently deploying work.
Leadership
Founders
Ari Wilson
Mathematics and computer science graduate of Vanderbilt University with an MS in computer science. His career includes 13.5 years at Google working on scaled frontends, APIs, distributed systems, and machine learning in ads; four years at Celonis launching Los Angeles and U.S. engineering and shipping process-intelligence products; and a four-month executive role defining agentic workflows at VideoAmp.
Founding Story
Wilson describes a long-running pattern of building side projects to his own requirements. After a VideoAmp layoff in August 2026, those side projects became his day job: he began building Bakeoff to evaluate coding agents on real bugs and features from his own codebases, with hidden tests and each agent running in its own harness. RemainderBot originated from his desire to use Claude and Codex subscription quota that would otherwise expire unused, turning it into reviewable work in private repositories.
Business Model
Revenue Model
No public revenue model is stated for Wilson personally. Bakeoff is described as a startup and offers 30-minute exploratory calls for teams considering agent evaluations, but no paid plan or price is published. RemainderBot uses the user's existing Claude/Codex subscription quota and does not require API keys or extra usage billing; the software is open source under MIT, with no public price stated.
Target Markets
- Companies selecting coding tools and agentic engineering workflows
- Engineering teams with proprietary codebases and test suites
- Individual developers and founders who have unused Claude or Codex subscription quota
- Executives seeking network-based job opportunities
- Individuals seeking a post-Mint personal budgeting tool
- Selecting coding agents for a company's own software and agentic engineering workflows.
- Evaluating bug fixing and feature development on proprietary code that public benchmarks do not represent.
- Using otherwise-expiring Claude or Codex subscription quota to complete delegated engineering, research, documentation, and follow-up tasks.
- Personal budgeting after Mint's shutdown.
- Executive job search and network-based opportunity discovery.