HeyJaredShip With AI
AI tools and automation, tested by a developer — not hyped. I build real workflows, connect real APIs, and show what they actually cost.
- Indexed videos, last 90 days
- 13
- Latest publication
- Sep 25, 2026
- Audience
- ~7 subscribers
- Earliest in this view
- Jul 23, 2026
Latest videos
A Company Actually Doubled Output With AI (opens the original)
Read excerpt
The most credible field studies of AI coding tools top out near a 40% gain. One company reached 209% - and let researchers into its telemetry to show how. 802 developers, 196,212 pull requests, 28 months. The doubling is real. It also took nine months, showed up almost entirely in newer repositories, and moved work downstream instead of removing it. Here is what did and did not explain the gain, the three limits of the study, and the one ratio you can track on your own projects this month. 00:00
`Why your worst AI bugs never get fixed (opens the original)
Read excerpt
A study of 300,000 AI-written commits found that 22.7% of the problems introduced are still sitting in those codebases. What decides how long a problem survives is not how bad it is - it's whether an automated tool complains about it. Full breakdown with five real repositories: https://youtu.be/Lg2NLG64Zac Source: arxiv.org/abs/2603.28592 #AICoding #TechnicalDebt #SoftwareEngineering ai coding ai generated code technical debt code quality linting static analysis software engineering developer sh
Nobody Ever Fixed These 5 AI Bugs (opens the original)
Read excerpt
A study of 302,000 AI-written commits across 6,299 repositories found that 22.7% of the problems the assistants introduced are still in those codebases today. What decides how long a problem survives is not how severe it is - it's whether an automated tool complains about it. Five real cases from public repositories, ordered by how long each one lasted, plus the five-minute check that tells you your own number. No new tools required. 00:00 - A request with no timeout, written in 2024 00:32 - 1.
5 Tests AI Writes That Prove Nothing (opens the original)
Read excerpt
A test suite with 100% coverage caught 4% of deliberate bugs. Coverage measures which lines ran — not whether anything was checked. Five failure patterns that show up in AI-generated tests, what each one looks like in code, and the two-minute measurement that tells you which of your tests actually prove something. No tools required. 00:00 - 100% coverage, 4% of bugs caught 00:25 - 1. Assertion theater 01:00 - 2. The mirror test 01:32 - 3. The mock that tests itself 02:01 - 4. Boundary blindness
5 Reasons Parallel AI Agents Don't Make You 3x Faster (opens the original)
Read excerpt
Three agents finished in the time one used to take, and your day didn't get three times shorter. The work went parallel; the finishing didn't. Five reasons the speed disappears, each with the counter-move that gets it back: no isolation means collisions, the merge queue is serial, review capacity is fixed, context switching costs you the attention you were going to review with, and an unattended wrong agent keeps going. Then the number that actually matters — how many agents your setup can absor
Publishing over time
Last 90 days. Choose a month to open its work.
Recurring subjects
Named in the text we hold. One piece can cover several.
Audience
~7 subscribers
Measured Sep 19, 2026
Source's subscribers, not the number who saw an individual piece.
About this data
Counts cover the work we have indexed. Tone needs enough text and a confident classification. Excerpts and episode notes are not full articles or transcripts.
Identity or attribution wrong? Suggest a correction.