Just ran Termyte’s local shell benchmark.
It blocked 96.4% of dangerous agent actions before execution.
The misses were still logged into local operational memory, so future sessions can learn from them instead of repeating the same failure pattern.
Not a sandbox. Not perfect yet.
But this is the direction:
AI agents can act.
Termyte makes sure those actions are governed.
One thing I wish the hiring industry would improve is respect for candidates' time.
Over the last few months, I've experienced interview processes that stretched for weeks and sometimes months. Candidates spend hours preparing, completing assignments, attending multiple rounds, and following up professionally.
Getting rejected is completely fair. Every company has a hiring bar.
What's difficult is when candidates are left without updates, get ghosted after multiple rounds, or encounter interviews where the interviewer is distracted, takes long breaks, joins late, leaves midway, or appears unprepared.
An interview is a two-way evaluation. Candidates are expected to show up prepared and take the process seriously. The same standard should apply on the other side as well.
A simple rejection email takes less than a minute to send. Giving someone your full attention for 30–60 minutes costs very little compared to the effort they put into preparing.
Candidate experience isn't just about hiring. It's about professionalism, respect, and recognizing that behind every application is a person investing their time, energy, and hopes into an opportunity.
Termyte just endorsed .agent — keeping the agentic web open, not owned by one company. https://t.co/8dQsM2P4P2 @agentcommunity_ https://t.co/8dQsM2P4P2
we’re hiring 10 Masturbation Consultants
$2,000/month to test our new Daily Guided Masturbation feature and document the effects on stress, sleep and mood
yes it’s real
yes you get paid
@ritu_twts One major thing is that most people were using supabase before starting to vibecode so they got used to it, and they have very generous free tiers.
Things I learnt this week building my startup:
-Don’t overcomplicate the product early.
-Warm intros beat cold outreach.
-Set clear goals instead of just building blindly.
-Talk to your ICPs before making product decisions.
Most startup mistakes are just building too much before understanding enough.
Every week I see another AI CEO saying software engineering is basically over in 6 months.
Then I open their careers page and they’re hiring:
distributed systems engineers,
infra engineers,
security engineers,
compiler engineers,
reliability engineers.
If AI was truly replacing engineers end-to-end that fast, why do the people building the future still need so many of them?