Sébastien Le Goff

GTM engineer building Walego. Mostly posting experiments: which model sorts leads, what it costs, where it breaks.

[Sébastien Le Goff @legoffseb]

Swapped Gemini for @typesafeai's Jev on the step that sorts contacts in Walego.

8,337 contacts in 73 minutes. Model bill: $0.32.

The part that almost broke it: asking Jev if it had enough data. It said no to everything. https://t.co/jDn6v0PfHU

[Sébastien Le Goff @legoffseb]

Swapped Gemini for @typesafeai's Jev on the step that sorts contacts in Walego.

8,337 contacts in 73 minutes. Model bill: $0.32.

The part that almost broke it: asking Jev if it had enough data. It said no to everything. https://t.co/jDn6v0PfHU

[Sébastien Le Goff @legoffseb]

@typesafeai @deel Another receipt for the pile. Swapped Gemini for Jev on the step that sorts contacts in Walego: 8,337 contacts in 73 minutes, $0.32 of model cost, zero errors. Wrote up what worked and the one trap I hit: https://t.co/WCllk3FfcY

[Sébastien Le Goff @legoffseb]

Swapped Gemini for @typesafeai's Jev on the step that sorts contacts in Walego.

8,337 contacts in 73 minutes. Model bill: $0.32.

The part that almost broke it: asking Jev if it had enough data. It said no to everything. https://t.co/jDn6v0PfHU

[Sébastien Le Goff @legoffseb]

@wandermist Test before you trust is the right call. My first test looked bad: 21 out of 60 matches against ICPs I'd already stored. Most misses were my own labels needing a company size that a LinkedIn headline doesn't carry. What did your test set look like?

[Sébastien Le Goff @legoffseb]

@hooeem One thing that bit me doing the same kind of ranking. I asked Jev a separate yes/no, "is there enough data to decide?", and it came back around 0.2 on everything, even clean inputs. Putting "undetermined" in the list of choices fixed it. Might matter for catalysts with thin info.

[Sébastien Le Goff @legoffseb]

@gippp69 Priced it on a real list before believing the 400x. 8,337 contacts sorted into 12 buckets, $0.32 of model cost. Same contact on Gemini 2.5 Flash vs Jev came out about 35x cheaper, not 400x. Did your desk lanes land closer to 35 or 400?

[Sébastien Le Goff @legoffseb]

@bageriarah @treg_ai @clay Can you share 2 or 3 enrichments you actually found useful? And roughly what those cost you on treg vs what the same thing would have run in Clay.

[Sébastien Le Goff @legoffseb]

@TheAwesomeTuvia nice one. i should definitely do this too

[Sébastien Le Goff @legoffseb]

Cold email doesn't die here, it gets a better filter.

Spam filters were supposed to clean up email and what they produced was deliverability as a full time job. Whatever the agent ranks on is the next thing everyone optimises for.

What actually dies is volume to the wrong people.

[GREG ISENBERG @gregisenberg]

I think cold email is going to die.

Every email inbox will have an agent gatekeeper soon, and the only way through will be a warm intro or being interesting enough that the agent decides you're worth its person's time.

[Sébastien Le Goff @legoffseb]

@zackkanter Probably, though the filter doesn't disappear, it moves. Spam filters were meant to clean up email and instead they created deliverability as a job.

Whatever an agent ranks on is the next thing people optimise for.

[Sébastien Le Goff @legoffseb]

@BenyaminHolley Not sure more training or context is what gets it there. What moved it for me was starting to build graphs.

Once each decision sits in its own node instead of one big prompt, the behaviour gets predictable. That's when it started holding up.

[Sébastien Le Goff @legoffseb]

Interesting. I think the next one includes being able to talk to agents. Every inbox is going to be handled by one before long, so you'll need to know how to speak to them.

Precision in the email becomes a lot more important. The offer has to be clear enough that the agent decides to bring it up to its owner.

[Sébastien Le Goff @legoffseb]

Been building an infrastructure that covers most of that list. You say what you want, it builds it for you. Sourcing, enrichment, deliverability, the orchestration layer, the Clay part.

AI is really scratching at this right now, especially with the graphs and loops that just came out. In a few years I think having the theory will still be worth a lot. The execution on top of it is going to be handled by agents.

[Sébastien Le Goff @legoffseb]

Best sales advice I've read anywhere is a Hacker News comment from May 2020. Someone asked how to brand yourself as a freelancer. A guy called goatherders, 20 years building sales orgs, answered this:

1. Sales is a lot like golf. You can make it so complicated as to be impossible, or you can simply walk up and hit the ball. Walk up and hit the ball.

2. Sales is about people and about problem solving. Not solutions or technology or chemicals or lines of code or artichokes.

3. People buy 4 things and 4 things only. Ever. Time, money, sex, and approval/peace of mind. Try selling something else and you fail.

4. People buy aspirin always. They buy vitamins occasionally and at unpredictable times. Sell aspirin.

5. All things being equal people buy from their friends. So make everything else equal, then go make a lot of friends.

6. Being valuable and useful is all you ever need to do to sell things. Help people out. Send interesting posts. Introduce people who would benefit from knowing each other, then get out of the way, expecting nothing in return.

7. No one cares about your quota, your payroll, your opex, your burn rate. They care about the problem you are solving for them.

Six years old and I haven't read better since.

[Sébastien Le Goff @legoffseb]

https://t.co/5H0qc3JwHQ deserves a spot in that conversation. A lot of what people call scraping is really just wanting Google's results as structured data, and that costs almost nothing through a search API.

Different job than Apify, but I hit it about as often inside the GTM agent I've built.

[Sébastien Le Goff @legoffseb]

@RCIllingworth Mozi

[Sébastien Le Goff @legoffseb]

@fivosaresti ABM

[Sébastien Le Goff @legoffseb]

@macomoroni what type of users are you looking for exactly and how do you find them ?

[Sébastien Le Goff @legoffseb]

@itsalexvacca The order is right. Step 1 is where I'd push: funding rounds, new VP, hiring spikes are the three signals everyone buys, so the accounts that fire are firing for your competitors too. The ones nobody sells yet sit in public records. Procurement awards, permits, registry filings.