
FDE at @ShieldedLabs, building @usemarkov and https://teapaw.org/t.co/1YZfMdqCeR. crypto x ai builder.
Joined October 2020
- Tweets6K
- Following1.8K
- Followers1.2K
- Likes5.7K
Pinned Tweet
copynpaste, 7 weeks in
- 6,300 people in the last 90 days
- 4 paying users
- 12 integrations shipped, every one from a user email
- $0 on marketing
built it because pasting between apps kept wrecking my formatting. turns out it wasn't just me
most developers still think zcash is just a coin. it's a whole privacy stack and almost nobody builds on it
what people mean by god changes from one country or community to the next. i don't believe in god the usual way, but i do believe there is an ultimate power out there. i can't explain it better than that
2
60
forward deployed engineer sounds fancy. it means go sit with the people using the thing and fix what actually hurts
1
1
59
i'll take a boring, well-documented codebase over a clever codebase any day.
clever code impresses people during a demo.
readable code saves everyone when something breaks at 2 am.
what matters more to you: cleverness or maintainability?
1
40
developer experience:
spend 20 minutes writing the code.
spend 2 hours understanding why the environment doesn't agree with you.
spend another hour fixing something that was working perfectly yesterday.
and somehow call it a productive day.
4
125
unpopular opinion:
we don't need more AI agents with access to everything.
we need agents that can do one task really well, with limited permissions, clear logs and no ability to quietly expand their own access.
autonomy without boundaries isn't intelligence.
it's an incident waiting to happen.
2
3
101
a bug in one Ethereum execution client can become a problem for multiple networks.
the recent Reth state-root issue led Base and BNB Chain to publish security updates.
this is why client diversity matters.
if every network depends on the same implementation, one bug can have a much bigger blast radius than expected.
source: crypto.news/bnb-chain-base-r…
2
69
a malicious Ruby package reportedly modified AI agent configuration files so the agent would execute it again.
read that twice.
the attack wasn't just stealing credentials during installation. it was trying to make the developer's own tools keep running the malware.
AI agents make developer workflows faster.
they also make malicious dependencies more interesting targets.
source: theinjection.dev/week/2026-w…
1
3
119
zcash just made a pretty meaningful change to private payments.
NU7 is live.
block targets went from 75 seconds to 25 seconds, and the upgrade also introduces limits on shielded transaction activity to reduce the work wallets need to do.
privacy infrastructure has to solve more than hiding transaction details.
it has to make private payments practical to use.
source: coindesk.com/tech/2026/11/06…
your documentation platform can become a security problem, too.
GitBook disclosed an incident involving an attacker reaching infrastructure credentials through AI agents.
integration tokens, API tokens and Git Sync secrets were among the potentially exposed data.
we keep securing production code while forgetting how much sensitive access lives in the tools surrounding it.
source: theinjection.dev/week/2026-w…
an AI agent reportedly went from exposed GitHub secrets to Azure Key Vault in just 5 minutes.
that's the part of agentic AI security i think deserves more attention.
the risk isn't only what an agent can generate.
it's what an attacker can make that agent access using credentials it discovers.
if your secrets are reachable, your permissions become the real attack surface.
source: theinjection.dev/week/2026-w…
your AI coding assistant might be executing code before you've even approved the repository.
a vulnerability in Mistral Vibe allowed crafted git configuration to run commands before the trust prompt appeared.
this is why "i trust the AI tool" isn't enough.
you also need to trust the repo, its config, its dependencies and everything it executes.
source:
theinjection.dev/week/2026-w…
do you still copy stuff out of chatgpt into slack and fix the formatting by hand or is that just me
1
4
70
