Hi, I'm Tom.
29, I live in San Francisco, and February is when I came on board at @looplyhqcom, as its co-founder and CTO.
The first thing I needed to figure out here: how much should outbound software decide on its own before a real person says yes?
You'd expect a tool like this to run by itself. Ours stops for a human at the steps that matter.
Where the checkpoints sit:
- The lead: 1 person and 1 dated signal, like "recently changed jobs" or "engaged with a post about X", scored against your ideal customer before a single word gets written
- The first email: AI drafts it from that 1 signal. You approve each one, or you choose to let the campaign run
- The inboxes: email goes out from your own Outlook or SMTP inboxes, with a daily cap on each one. Gmail inboxes only send to people who opted in
- The LinkedIn side: 20 invites and 40 messages a day by default, Monday to Friday, from your own account, and every message can be reviewed before it goes
- The replies: 1 shared inbox where AI drafts an answer and someone reads it before it's sent
Put together: the signal picks who, the score decides whether, and a person signs off on what gets said. Email and LinkedIn sit in one sequence, and the LinkedIn steps wait for an accepted invite or a reply before moving on. That's how it runs today, and it's how we run our own cold email too.
The bugs that worry me most here aren't the complicated ones. They're the small ones that could land on a paying customer, with nobody above me to escalate to.
My biggest takeaway as CTO: a decision that looks smart today is only smart if it doesn't slow the whole team down six months from now.
@antinertia the role is the useful bit here, a company hiring an SDR has a different problem than one hiring a data engineer, so the opener should change with it
After six guides in three weeks, this is the model I'd use for each outbound job. Save it:
Scoring a batch you want to see and move around → Claude Opus 5.5
It builds the scorecard right inside the chat, with sliders, cards and the evidence next to every score
Writing the one email that matters → Claude Opus 5.5
Judgment, plus the patience to leave out what it can't prove
Sorting replies → any of them, as long as it goes by the latest message and puts unsubscribes first
The rule under all of it: use whichever model is best at the job. Last week's pick gets no free pass
And the part no model does for you: opening the source and checking the company