Replying to @wiz_io
@wiz_io research uncovered PHP2Shell, a serious @php_net vulnerability.
Could the @startearly_ai Release Guard have flagged it when introduced in January 2025?
We replayed the change. Early flagged the unsafe memory lifetime and created a Fix Pull Request.
Run Early nightly.
1
24
56% of estimated engineering effort went to repairs in our study of four open-source projects.
That’s why this launch matters to me. Early now finds bugs and creates fix PRs with tests.
Less repairing. More building.
startearly.ai/
Find bugs. Create fixes. Get back to building.
Early finds bugs and creates fix PRs with tests. Your engineers review and control the merge. Less repair overhead. More capacity to build what matters.
My completely non-scientific AI model benchmark: if the agent using this model were an employee on my team, how much would I trust its work vs. inspect every deliverable?
Based on work across product, development, content, sales and marketing.
Probably obsolete next month :)
18
Open source is dead. Long live open source.
I moved our website and docs into one GitHub repository. Astro powers the site. Pagefind powers documentation search.
Open source feels like LEGO now: choose the pieces, connect them, and own the system.
startearly.ai/docs/
1
20
The full founder story, including why I replaced Webflow and Docusaurus:
startearly.ai/post/i-turned-…
1
16
The open-source building blocks:
Astro for the website:
astro.build/
Pagefind for documentation search:
pagefind.app/
5
A fix can work and still break something else.
We replayed a real @mastra regression with @startearly_ai's Regression Guard. None of its GitHub checks failed, but Early traced the other behavior that changed.
Technical replay: startearly.ai/regression-cas…
Know a regression? Send it our way.
13
My contribution to developers today: a practical five-gate workflow for Jest tests with AI coding agents.
Inspect context. Plan around risk. Generate one focused test. Challenge its assertions. Finish with evidence, not confidence.
Prompts included:
dev.to/sharonyb/jest-unit-te…
2
2
43
Fall in love with the problem, not the solution.”
Uri Levine’s title still resonates with me.
The problem I’m focused on isn’t reviewing new code. It’s asking what a release could break in a system that already works.
Code review ≠ regression review.
linkedin.com/pulse/code-can-…
12
Two weeks with OpenAI Codex (5.6 Sol, Fast mode), and I have the same feeling I had when I first tried Claude: a serious step up.
I now use it across product, development, marketing, and operations. I rebuilt startearly.ai from scratch.
Same team. Stronger body.
Do you feel the same?
🤖 Made with AI
22
We’re testing Regression Guard against known regressions in open-source projects.
First: Pydantic AI. A 47-file fix passed 78 CI checks, but one removed loop could cause incomplete streamed responses across 14 releases.
Here’s the independent replay:
startearly.ai/regression-cas…
35
Claude Code is down. Very bad.
This weekend I completed work that, just 4 months ago, would have taken an entire scrum team a month to finish. I was about to start what would’ve been the second month of work… until this happened.
Is it apparent that i got addicted to Claude 😂 ?
2
3
486
Check out my latest article: Pull Request Coverage: A Practical Way to Measure Quality in PRs linkedin.com/pulse/pull-requ… via @LinkedIn
2
27
The Next Frontier: Autonomous Agents for Test Code Generation
linkedin.com/pulse/next-fron…
16
Sharon Barr retweeted
Think TDD is dead?
It’s not. It’s evolving.
TDD was never about tests — it was about confidence.
Now AI writes the tests. So the flow changes:
1️⃣ Describe intent
2️⃣ Agent writes the failing tests
3️⃣ You implement
4️⃣ Regenerate + iterate
Test-Driven becomes Agent-Driven.
Is AI changing how you test?
#TDD #AIforDevTools #EarlyAI #GenAI #DevTools #SoftwareEngineering
Check out my latest article: Collaborative AI Agents in Software Testing: The Developer Path to Code Quality linkedin.com/pulse/collabora… via @LinkedIn
64
How do you distill a year of hard work into 60 seconds?
As a startup CEO, storytelling is the ultimate skill. You juggle roles like product manager, sales lead, and cheerleader, but storytelling stands out as the most important.
At ICON Israel collaboration network, we learned how to structure a great startup story:
1️⃣ Team
2️⃣ Problem
3️⃣ Why now?
4️⃣ Solution
5️⃣ Differentiation
6️⃣ Market
Telling this in 5 minutes is hard—condensing it into 1 minute is even harder! Last week, with help from @Donna_Grif , I delivered my 60-second pitch to ICON’s VCs, execs, and community.
Here it is: a year of hard work in just one minute. 😊
Check out my latest article: From Mountain Peaks to Startup Milestones linkedin.com/pulse/from-moun…
78
Here comes @vitest_dev support by @startearly_ai !
Thomas Dohmke, CEO of GitHub, says: “50% of all code produced by Copilot users is now AI-generated.”
But… who’s testing this code? That’s where Early comes in—an AI Agent for test code.
Today, we’re adding Vitest to our supported frameworks! Generate mocks, happy paths, edge cases, and high-coverage tests with ease.
We ran Early on Mock Service Worker (MSW) by @kettanaito:
📊 637 tests generated
✅ 458 green
❌ 179 red
Early is free for OSS—happy to collaborate with MSW. Cleaning tests & sharing soon!
More frameworks, features, and Early magic coming.







