Kash Sajadi

Eclectic, but not exclusively. CEO @cloud66.
Principal partner at AltaCoda Venture Studio:

[Kash Sajadi @khash]

what's the difference between skill and experience?
why would you put "CS degree with 5 years of experience" in a job description?
agent "skills" and CLAUDE (sorry AGENT).md files are just that: skills. they are not experience.experience comes from learning by doing.

[Kash Sajadi @khash]

@Nick_333ch stay on course! owning the ses account is the right base, the bill stays the ses bill. the hard part later is deliverability: dmarc, reputation, warming. we went down the same road with sendops, happy to compare notes.

[Kash Sajadi @khash]

@wrenchtokeys ouch, silent drops since august is the worst kind of failure. glad the agents caught it. ses suppression is easy to miss because nothing errors on your side. that's why we put bounce and reputation monitoring into sendops, on your own ses. https://t.co/1XIlBJcmyz

[Kash Sajadi @khash]

@shopkeeperseo i think going direct to ses is worth it once volume grows, the send cost is hard to beat. the part you end up rebuilding is drips, bounces and reputation. we built sendops for exactly that: workflows on your own ses, flat pricing, no per-contact fees. https://t.co/1XIlBJcmyz

[Kash Sajadi @khash]

@byrehan @dannypostma good list. one box for many side projects is the right instinct, the bill stops growing with every thing you launch. if you want the git push feel without running the paas yourself, that's what cloud 66 does on your own servers. https://t.co/c8wZjTaCOL

[Kash Sajadi @khash]

@ksaitor @HellShowers @vercel @CryptoJobsList fair points, next.js builds are hungry. one thing that helps: build somewhere other than the box serving traffic, so an 8gb build doesn't take the site down with it. if you ever want the vercel-style git push on your own vps, that's what cloud 66 does.

[Kash Sajadi @khash]

@marcos_placona haha! you won't get sued, most of us have been there. the bit people underestimate is backups, db upgrades and ssl, not the first deploy. that's the gap cloud 66 fills: git push to your own servers, managed databases and backups, full ssh. https://t.co/c8wZjTaCOL

[Kash Sajadi @khash]

@PowerduckLabs the npm one is painfully familiar. nobody owns the memory file, so it's always a few months behind. we flipped it in bench47: agents write back what changed when they finish, so the next one doesn't start from last spring. https://t.co/Bgabv1aRSq

[Kash Sajadi @khash]

@theomlondhe great data. 13 maintaining two different files is the scary number, that's two sets of rules quietly drifting. i'd go pointer or symlink, and keep the stuff that changes often (decisions, past incidents) out of both. that's what bench47 is for.

[Kash Sajadi @khash]

ha, fair for a lot of what's out there. to me it's a lot more than building fast or building automated: it is about the difference between skill and experience.
you can think about it this way: all CS grads have skills, but they gain experience as they work. the MD files (and agent "skills") as skills. experience comes from learning from doing.
that's the part bench47 does. https://t.co/8ykRYejFCH

[Kash Sajadi @khash]

@AhmedReza one source generating both is the right call. i think the drift that hurts more isn't between the two files though, it's between the files and what the team decided last week. we keep that part in bench47 so claude and codex start from the same decisions. https://t.co/Bgabv1aRSq

[Kash Sajadi @khash]

honestly not well, as long as the lesson lives inside one tool's memory. what worked for us is keeping it outside every tool: decisions and incidents in one shared place each agent reads before it starts and writes back to after. that's what we built bench47 for. https://t.co/8ykRYejFCH

[Kash Sajadi @khash]

most onboarding emails wait on the clock. the better signal is what the person did. in sendops a workflow can wait for a first project, then send the next tip, or offer help if two days pass without one.
https://t.co/1XIlBJcmyz

[Kash Sajadi @khash]

AWS Nitro Sandboxes gives each EKS pod or ECS task its own hardware-isolated VM, at no extra charge. Hard to be a standalone agent sandbox company when the cloud makes isolation a one-line change.
https://t.co/sVZTKsMWqw

[Kash Sajadi @khash]

@iamsrikanth_dev stay on course! 4 of 5 stuck on the same step is a great signal. an email that only goes to people stuck on the api key, with the exact fix, will probably do more than the welcome email. that's the kind of flow sendops runs on your own ses. https://t.co/1XIlBJcmyz

[Kash Sajadi @khash]

@nikhilx22 amen! and the ones that work are usually behavioral, not a timed sequence: nudge the people who signed up but didn't do the first thing. sendops runs that on your own aws ses if you're already there. https://t.co/1XIlBJcmyz

[Kash Sajadi @khash]

@imvincz @levelsio $450 to $12 is the usual ses story. the part people end up rebuilding is drips, bounces and reputation. we built sendops for exactly that: workflows on your own ses, flat pricing, no per-contact fees. https://t.co/1XIlBJcmyz

[Kash Sajadi @khash]

@bbbrint same. kubernetes is great when you need it, and most apps don't. a few servers with good deploys, backups and ssh covers a lot. that's been our bet with cloud 66 for years.

[Kash Sajadi @khash]

@vltg__ i am always weary of free tiers. they are marketing loss leaders but a small vps is a few bucks a month and doesn't disappear on you. if your friend wants the heroku git push feel on it, that's what cloud 66 does. https://t.co/c8wZjTaCOL

[Kash Sajadi @khash]

@AndreaChello paying for an instance to sit there waiting on an llm is the worst part of that model. your own servers don't care how long a request waits. cloud 66 does git-push deploys to your own cloud, so idle time stops being a line item. https://t.co/c8wZjTaCOL