Pinned Tweet
Plotted the cost of building software since 1980, Moore's law style. Halves every 10 years, then every 2 from 2022.
What did I miss?
What did I miss?
0
0
0
31
Your AGENTS.md steers your coding agent more than most of your code, and almost nobody reviews it.
A planted one in a Downloads folder got Codex to install malware in over 90% of Forever Security's tests. Nothing to patch: the agent did what the file said.
A planted one in a Downloads folder got Codex to install malware in over 90% of Forever Security's tests. Nothing to patch: the agent did what the file said.
1
0
0
36
The write-up, including the Claude Code test where a planted CLAUDE.md got a risky settings write approved about half the time: thehackernews.com/2026/10/malici…
0
0
0
15
Sweden's government blocked the Claude and ChatGPT websites on staff PCs today. Staff still get both models through an internal platform with a contract behind it.
Same question for your coding agents: whose account does your code go through, and who set its data retention?
Same question for your coding agents: whose account does your code go through, and who set its data retention?
2
0
2
65
Your AI Agent Just Shipped a Hardcoded Key. Again. albert515763.substack.com/p/your-ai-agen…
0
0
0
24
Albert Franquesa retweeted
@albertfranquesa just published Your AI Wrote the Code. Who Checked It? medium.com/p/your-ai-wrot…
0
1
2
54
Love it when @AnthropicAI accepts your MCP into the store with a top-tier rating for @claude_code
If you're a developer building in the Claude ecosystem, you need to check this out. Norma by Quality Clouds is live: claude.ai/directory/norm…
If you're a developer building in the Claude ecosystem, you need to check this out. Norma by Quality Clouds is live: claude.ai/directory/norm…
0
0
0
23
My agent runs Norma before it opens a PR
It calls the MCP server, gets a verdict on the code and every call lands in the audit trail
Same code, same verdict, every run. And a record that exists outside the model
That's governance inside the agent loop.
norma.qualityclouds.com
It calls the MCP server, gets a verdict on the code and every call lands in the audit trail
Same code, same verdict, every run. And a record that exists outside the model
That's governance inside the agent loop.
norma.qualityclouds.com
0
0
0
24
4/ The generator changes the density by 2x.
Bolt and Copilot-assisted repos produce about half the issue density of v0 and Lovable, and a third to a half of their security density.
99% of Lovable projects carry a security finding. Two of 183 are clean.
Bolt and Copilot-assisted repos produce about half the issue density of v0 and Lovable, and a third to a half of their security density.
99% of Lovable projects carry a security finding. Two of 183 are clean.
2
0
0
12
Caveats, because they decide what this is worth:
Copilot-assisted repos are a control group. A human drove the architecture. 10 of our 295 rules produce 88% of findings. v0 against Lovable at 3rd and 4th is weakly supported.
Data and corrections: github.com/qualityclouds/…
Copilot-assisted repos are a control group. A human drove the architecture. 10 of our 295 rules produce 88% of findings. v0 against Lovable at 3rd and 4th is weakly supported.
Data and corrections: github.com/qualityclouds/…
0
0
0
9
We scanned 429 AI-built apps. Here's what the code actually looks like. open.substack.com/pub/albert5157…
0
0
0
35
Albert Franquesa retweeted
I just published "Your AI Wrote the Code. Who Checked It?" medium.com/p/your-ai-wrot…
0
1
0
19
