
Co-Founder: @OrbitalBeam Access & Distribution for ai, blockchain & emerging tech
San Francisco, CA
Joined February 2012
- Tweets27.8K
- Following6.4K
- Followers39.5K
- Likes30K
Pinned Tweet
privacy is for everyone
privacy is for your identity
privacy is for your pii
privacy is for your financials
privacy is for your transactions
privacy is for your health
privacy is for your data
privacy is YOUR choice
privacy is for _____ ?
Cursor just pulled a Claude Code extension that Anthropic never built. It spent ~11 hours in Cursor's marketplace trying to backdoor dev machines over SSH.
What happened: v2.1.313, not built by Anthropic, tried to plant an SSH backdoor (Cursor security notice).
My take: your extension store is supply chain now. Pin versions.
Why it matters: coding extensions run with your dev access, so one poisoned update lands right where the keys live.
Source: Cursor's security notice (check for the svc-telemetry key + remediation scripts) cursor.com/blog/claude-code-…
Hot Take: the first big enterprise agent disaster will have a human sign-off on every step.
"Human-in-the-loop" sounds safe … until the human approves 400 actions a day between meetings.
That's basically a rubber stamp with a calendar invite.
The agent didn't go rogue. Accountability just went to lunch!
Everyone's worried AI agents will rob banks for themselves.
The likelier version is way more boring … we hand them the keys ourselves.
Withdrawal rights & no spend cap.
Bank robbers used to dig tunnels. Now they wait for a permissions screen nobody read.
One startup has 15 employees and ~150 AI agents.
That’s a hell of an org chart.
But if you’re measuring agent headcount, you’ve already picked the wrong KPI.
Show me work completed, rework, cost per outcome, and customer trust.
Nobody ever won a market by hiring the most bots.
OpenAI just raised prices without touching the price.
- Codex + Work: 20x Plus → 10x
- GPT-6 Pro: 200 msgs/week → 100
- Price: still $200
The real signal…
flat-rate AI was a customer acquisition cost, never a pricing model.
The free trial just ended.
Update: the free trial got extended.
Switching costs in AI are one browser tab wide ... so the customer sets the price - live, in the replies
xitter.gq/thsottiaux/status/2115…
Day 25/
We are doubling all usage limits for all Pro 100/200/500 tiers and Business Premium, effective today.
We heard your kind feedback and won’t be reducing usage limits for Pro 200, so your Work and Codex usage will stay the same. We’re also going to continue offering extended limits for dots for now, and usage of your dot will still continue to be separate from your default usage.
And we'll of course also keep shipping better models and other efficiency improvements that increase your effective usage over time.
I'm starting to sense a divide in AI rollouts.
1. Teams counting what agents produce
2. Teams counting what a human checked
The first group has amazing dashboards … while the second knows what shipped.
More cooks don't help if one guy still tastes every plate.
ICYMI: Alphabet just reported its fastest search query growth in five years and still missed on search revenue. Shares fell 13% this morning (FT).
Good News: the usage showed up.
Bad News: the money lagged.
Do we blame the agents? 👇
What happened (Q3 earnings report):
- Search +12% to $63.35B vs $64.96B est., 2nd straight miss
- Cloud +103% to $30.7B
- Free cash flow -$9.07B
- 2026 capex held at $195-205B, set to "increase significantly" in 2027
Why it matters: Alphabet blamed FX first. The FT reads it as AI Mode pressuring ad profits.
My take: owning the front door only pays if the toll booth moves with the traffic. Search now has to learn how to bill an agent.
Source: Q3 report via CNBC cnbc.com/2026/10/28/alphabet…
4:45am. Coffee & a moon.
The agents worked the night shift … I'm just here to approve the overtime
Best meeting of the day has zero attendees
I’ve seen a ton of great hacks for TVs lately, getting rid of bloatware and setting up a better ui.
Anyone done the same for Roku devices?
Most "AI-powered workflows" are still the same old workflow with a chatbot taped to the front.
Real agentic execution flips the model.
Human defines the outcome.
Agent figures out the steps.
Enterprise verifies the result.
If your AI still needs someone clicking through five dashboards, you've automated the receptionist.
Not the business.
Google just took the #1 spot on the Artificial Analysis index with Gemini 4 Carbon, and almost nobody can use it yet.
What happened (Google blog post):
- #1 on Artificial Analysis Intelligence Index v5.0, at 59
- SEC-bench Pro 88.2%, ExploitGym 54.8%, CWE-bench 71.0%
- This week: some cyber defenders only (Fairwind Program); devs later
- $2/$10 per 1M tokens intro, then $4/$20
Why it matters: Google's best model ships with a public price card and a closed door, hours before Alphabet earnings.
My take: gated access is GTM now. Defenders first, safety story banked, everyone else waits. Velvet rope, metered bar.
Source: blog.google/innovation-and-a…
Anthropic just swapped its $1 government seat for a meter. Smart trade.
What happened (blog post below):
- 40% off Claude, every surface
- Federal to tribal
- No minimums, pay per token
Why it matters: $1 deal ends Oct 31.
My take: the $1 seat was the sample. Now the meter runs.
Source: Anthropic blog post, plus GSA's release. anthropic.com/news/expanding…

