How I Get 8M+ Leads Using Claude Code (For Cold Email)
A cold-email agency swapped a $25K/month Clay stack for two free Claude Code skills that process leads locally, no row caps or seat limits required.
Posted
yesterday
Duration
Format
Tutorial
educational
Views
4.8K
61 likes
57 · 43
Big Idea
The argument in one line.
A cold-email agency replaced a $25,000-a-month SaaS and infrastructure stack with two Claude Code skills, one a plain-language interview and one pure deterministic code, cutting the same processing work down to a $200 subscription.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
You run cold email, LinkedIn outreach, or any list-building process and are paying per-row or per-seat for a SaaS enrichment tool.
You're comfortable giving an agent API keys and letting it run a multi-step workflow with no UI in the loop.
You want a concrete example of splitting an agent workflow into a language skill and a code skill instead of one long prompt.
SKIP IF…
You want a plug-and-play tool today. The two skills mentioned still require API keys, a data provider, and setup work before they run.
You want a personalization playbook. This video argues personalization stopped moving the needle for their campaigns, it isn't advice on writing personalized emails.
TL;DR
The full version, fast.
A cold-email agency used to spend $25,000 a month on Clay and lead data before hitting Clay's row caps and seat limits at 17.3 million weekly enrichments. They rebuilt the pipeline as custom code, first on a cloud system processing a quarter-million rows a second, then simplified it down to two free Claude Code skills that run on a laptop: one interviews you for your ideal customer profile in plain language, the other runs deterministic code to search, verify, and clean the list. The team also stopped personalizing emails and stopped scoring leads, favoring one well-targeted send to their whole addressable market every 30 to 60 days over complex sequences.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Cold open: a teammate processed 8 million leads on a laptop with no SaaS. The old stack cost $25K/month; the new one is a $200 Claude Code subscription.
00:39 – 02:06
02 · Why Clay broke for them
The standard SaaS stack (Clay + tables) caps work by what you're willing to pay, not what you need. At 17.3 million weekly enrichments they became Clay's largest user and hit every limit.
02:06 – 03:05
03 · Rebuilding the waterfall in code
They rebuilt Clay's function as custom code: a cloud system doing a quarter-million rows/second versus Clay's ~27 hours per million. It's now simple enough to run on local MacBooks with just Claude Code.
03:05 – 04:42
04 · The two skills: interview, then pure code
Skill one is pure language: it interviews you on target, list size, and requirements, then returns a plan before touching data. Skill two is pure code: dedupe, strip role addresses, format, because row-by-row AI decisions wreck context windows.
04:42 – 07:38
05 · Context in, waterfall out
Real client onboarding uses a campaign ideation doc to define ICP and TAM up front. The search step then runs a waterfall across data providers like MoltSets, cascading to the next provider on any invalid result.
07:38 – 09:54
06 · The output and the export
Example run: 380,000 sales leaders filtered by company headcount and sales-department size. Export is the awkward, custom part: their NDA'd partner Halo needs manual 5,000-row chunk uploads; most viewers can just use an API like Instantly or Smartlead.
09:54 – 12:19
07 · No personalization, no scoring
The same list works for LinkedIn, ads, and calls, not just email. Personalization stopped moving results about six months ago; a zero-personalization RB2B campaign still generates six figures a month. They don't score leads either, they just email the whole TAM every 30-60 days.
12:19 – 14:58
08 · Running the company on agents
A rotating pool of 7 Claude and 4 Codex subscriptions powered 6,092 agent sessions and ~250K tool calls in a month. The real cost was $28,275 (Codex 54%, Claude 46%) thanks to heavy prompt caching, and the $200 plan only multiplies the 5-hour window, not weekly capacity. Ends on the CTA and a next-video tease.
Atomic Insights
Lines worth screenshotting.
A team replaced a $25,000-a-month Clay and data-provider stack with a $200 Claude Code subscription doing the same processing work.
Clay's row caps and seat limits didn't break because of a bug, they broke because the pricing model caps work by budget, not by need.
A custom code pipeline processed a quarter-million leads per second versus Clay's roughly 27 hours per million leads.
Splitting an agent workflow into a language-only interview skill and a separate pure-code execution skill avoids burning context on row-by-row decisions.
Letting an AI read a CSV one row at a time to make decisions is a context-window-killing anti-pattern; deterministic code should do that work instead.
A verified lead waterfall has three steps: search for the person, find their contact details, then validate the contact before trusting it.
When a data provider returns an invalid email, the workflow should automatically fall through to the next provider rather than accept bad data.
This agency stopped personalizing cold emails about six months ago after testing showed no measurable difference in reply rates.
A single generic RB2B campaign with zero personalization has generated hundreds of thousands of dollars a month in revenue.
Instead of scoring leads, this team emails their client's entire total addressable market once every 30 to 60 days to avoid burning the list.
The $200 Claude plan is not 20x the usage of the $100 plan overall, it only gives a 20x multiplier on the five-hour session window.
Running multiple cheaper Claude subscriptions and rotating between them can cut a company's AI bill roughly in half for the same total inference.
One month of agent-driven work at this company logged 6,092 AI agent sessions and about 250,000 tool calls from a single founder's workflow.
Takeaway
Local agents just replaced a $25K SaaS lead stack
AGENT WORKFLOWS
A cold-email agency swapped Clay and a custom cloud pipeline for two Claude Code skills running on a laptop, and the real lesson is in how they scoped the work, not the tool.
01The $25K teardown
A stack that once cost $25,000 a month in data and processing fees can now run on a $200 Claude Code subscription.
The bottleneck in lead generation isn't finding data, it's the software markup on processing it.
02Why Clay broke for them
SaaS lead tools cap you by what you're willing to pay, not by how much work you actually need done.
Row caps and seat limits aren't bugs, they're the business model, and they bite hardest once you outgrow the plan tiers.
Being a platform's largest customer is a warning sign you've out-scaled the tool, not a badge of honor.
03Rebuilding the waterfall in code
A quarter-million rows per second versus 27 hours per million rows is the real gap between custom code and a SaaS abstraction layer.
Once a workflow is simple enough, you don't need the cloud infrastructure you built to run it. A laptop is enough.
04The two skills: interview, then pure code
Separate the language layer from the execution layer: one skill interviews for intent, the other runs deterministic code, and neither should do the other's job.
Never let an agent 'think through' a CSV row by row. That burns context and produces inconsistent output; write code that transforms the file instead.
05Context in, waterfall out
The quality of an automated list-building run is set entirely by the context you hand the agent up front, not by the run itself.
A waterfall has three real steps: search for the person, find their contact details, and validate the result before you trust it.
When a data provider returns an invalid contact, cascade to the next provider automatically instead of accepting bad data.
06The output and the export
The last mile, actually delivering leads into a CRM or sequencer, is usually the most idiosyncratic part of the system and the hardest to templatize.
Most sequencer tools have an API for delivery; if yours doesn't, that's the one piece worth building custom around.
07No personalization, no scoring
Test whether personalization still moves your numbers before you keep paying for it. For this team it stopped working about six months ago with no measurable lift.
A single well-targeted email sent to 100% of a total addressable market outperformed a four-part personalized sequence sent to a scored subset.
The offer is still the entire game. No amount of research-backed personalization fixes a mediocre product.
08Running the company on agents
Actual API-list-price cost and subscription cost can differ by 10x or more once prompt caching kicks in at scale.
A $200 plan's advantage over a $100 plan is a 20x multiplier on the 5-hour usage window only, not overall weekly capacity, so multiple cheaper accounts can beat one expensive one.
Running dozens of parallel agent sessions is itself an operating model, not just a coding habit, once it's producing hundreds of PRs a month.
Glossary
Terms worth knowing.
Waterfall enrichment
A cascading process where a data provider is tried first for a contact, and if the result is invalid the workflow automatically moves to the next provider until one returns a verified contact.
Row cap
A SaaS pricing limit on how many records can exist in a single table or be processed in a billing period, forcing large-volume users to split work or pay more.
ICP (Ideal Customer Profile)
A defined description of the exact type of company and buyer a business wants to target, used to filter a raw lead list down to qualified prospects.
TAM (Total Addressable Market)
The full set of companies and buyers that could realistically become customers, used here as the target size for a single outreach campaign rather than a subset.
Prompt caching
A pricing mechanism where tokens repeated across requests are billed at a steep discount, which is why heavy agent usage can cost far less than the sum of each session's raw API price.
“Clay would take about 27 hours to process 1 million leads while at a quarter million per second we were able to do it in four to five seconds.”
hard performance comparison with a specific multiple→ IG reel cold open↗ Tweet quote
04:13
“Having an AI read one row of data, make a decision, and then go to the next row and then do that a million times over is horrible for your context windows.”
“I thought the $200 plan meant that you get 20x more usage, but that's not true. You only get a 20x multiplier on the five-hour usage.”
specific, actionable pricing correction viewers can act on immediately→ IG reel cold open↗ Tweet quote
The Script
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
17px
metaphoranalogy
Last week, just one person on my team uploaded around 8 million leads for our cold email campaigns. She did it on a 16 gigabyte MacBook Air sitting on her desk. without any of the software that most people use for lead generation.
So that means no clay, no Apollo, no tables, no row caps, no seat limits, literally just Claude code with fable 5 .1 running on her laptop. You see, we used to spend 25 grand a month just on lead data and the software to process it. And now the processing side of that set of costing 25 grand costs us a $200 a month Claude code subscription.
So in this video, I'll show you exactly how we are building lead lists. at scale, what we have had to replace to get there, and how you can run a similar system for yourself. So before I show you what we run now, you need to see what this used to require.
You see, the standard setup is a series of SaaS tools stacked on top of each other. How most people do it these days is you get a subscription with Clay, you load it up inside of a bunch of tables, and on their new pricing, they're going to charge you for every single row that you're doing, and it gets really, really expensive if somehow you aren't With row caps, like so you can't have more than like 50 ,000 rows on a single table.
So you have these table limits. Of course, you have like seat limits, things like that. And for most people, it's great.
For us, Clay stopped working for us about a year ago. We just started hitting all the limitations that are put into place on there for us. In that model, the amount of work that you want to be able to do is capped by what you're willing to pay, not by how much work you actually need to get done.
And that set up broke for us because it was never built for our volume. We used to be the largest user of Clay. I was on a call with one of their higher ups and he just...
jokingly laughed when i told them how much volume we were putting through there we were doing about 17 .3 million enrichments every single week on their platform and when i told that to this guy he jokingly was like wow i think you're using the platform wasn't literally anybody else and so we used to be one of the largest users of clay and i do want to be extra extra clear i think clay is a great product and i genuinely do believe most people should be using it this is just our experience it just didn't work for us anymore and so we had to rebuild the whole thing to run locally using cloud code anyways that brings us closer to today so we rebuilt what clay was doing for us we rebuilt the whole thing using custom code it was on a cloud and that system waterfaller would enrich about a quarter of a million rows per second and that's so fast it's genuinely unnecessary for most of what we do but for comparison clay would take about 27 hours to process 1 million leads while at a quarter million per second we were able to do it in you know four to five seconds now if a client needs their entire market pulled cleaned and ready to send in a day we are
actually able to do that on our side so the main thing i want you to take away is that you can process millions and millions of leads locally even on your own computer because since we did that waterfall thing it's got so simplified that right now we don't even use that cloud system anymore and we run everything on our local macbooks which is pretty dope and all you need is cloud code and a good data source so let me show you exactly what this looks like right now you see the whole thing is just two cloud codes and we will give those to you they'll be in the description for you and you just run one and then And so when you click that link, it'll bring you to this page.
I'm gonna update the copy here. So it might be a little bit different when you get this after we edit the video, but basically it's two primary skills that are going to be popping in here. And the reason that it's two separate skills is because this first skill is very much, it's just pure language.
There's no code in it at all. So all it's gonna do is tell your agent what our process is, which providers we use and in what order we use them. So the first thing it does is that it should interview you.
It should ask you who you want to talk to as in who you want to target, what size list you're trying to build and what your requirements are. Then it comes back with a plan before it touches anything. You can tell it to use different providers or reorder the waterfall and it will be able to adjust everything.
And so that first skill is very much to like prep the primary skill. So that way it can fit for your systems, what your data providers are or things like that. And the second skill here is where I think most people building this type of system go wrong.
You see the second skill is purely code. And so this one does a bunch of stuff for us. It uses code to strip out role addresses, like info sales, that it formats everything's correctly.
It'll D duplicate the list and the. is that all has to be code a lot of people will try to get their agents to try to like think through a csv file it's like no you actually need to run code on the file so you can get the exact outcomes that you want because having an ai read one row of data make a decision and then go to the next row and then do that a million times over is horrible for your context windows and that's exactly how we're able to generate and process millions of lead for so cheap and so that's where the first skill being language based is really important because it tees up the actual code run to make sure that you shouldn't run into any issues.
And so if you want these two skills yourself, the link here will be down in the description. And speaking of descriptions, if at any point in this video, you want to see how we could set up our entire system for you, enter cold email into a reliable lead generation machine for your business, click the top link in the description to book an intro call with us.
So now let me walk you through what actually happens when you run this entire skill from start to finish. And so it's going to ingest a raw CSV file. It does dynamic API query into search for the leads you asked for.
it runs that through waterfall enrichment to find and verify the contact details and then connects to your crm or your sequencer or whatever once again that first skill should interview you enough to make it so it can kind of fit into what your ecosystem is but maybe it's going to point out to you like hey you're actually missing this piece what is your sequencer you're missing um a data provider like blitz api or molt sets or whatever you're using and so in that step you're going to give it its context and the quality of the entire run is going to be decided there for example what we do is we give every single client we onboard a campaign ideation doc and during their on process with us they fill this all out and we're able to hop on a launch call with them and so instead of answering the interview questions with our agent we're able to just give them our clients campaign ideation docs and it's able to go through the entire thing because as you can see in here we are figuring out like hey what is your ideal customer profile and then we're also asking them like what are the soft requirements so we can actually define their total adjustable market and then zone in on what their ideal customer profile is as well so really if you have an icp doc a positioning doc or notes from past campaigns sitting in your business you can just link
that stuff directly as long as your agent can access it and to be able to pull everything through there and so step two in there is the searching it runs as a waterfall across all of our providers and so for example i was doing a test run yesterday with molt sense which will be releasing any day now i think we're on target to launching their campaigns in october 1st if you don't know what molt sets is it's a really good data provider from adam robinson we helped him launch our b2b about three years ago it's now a 9 .5 million company and our cold emails are responsible for 42 of their revenue.
And so he's launching multisets. I wanted to test it out. What's great is that none of these are apps that you need to log into or click around.
Your agent will know that they can access it through APIs. So you can just get a key from each provider and then Claude calls them directly from your machine. And it's not like the skill itself is doing the searching either.
It's just telling the agent which provider to hit first and then which one to come up after that and so on. And so the important parts to me in any waterfall are very straightforward. You have searching, you have finding and you have validating.
Searching here means something like find me everybody who's a manager. and above and marketing at XYZ types of companies. And there are obviously millions of combinations of potential search that you could do.
A search could even be a Google scrape or something like that. And then after that, once you search for the person, then you need to find the person's contact details. And that's where you go into a waterfall.
Because when you find that person's email, it's probably going to not be their actual email or it's invalid or bad data. So you need to run it through a validator to validate that what you got is good. And so for example, this skill, it would go to molt sets.
It would get an email. We'd go cool. Molt says that's the email.
Let's validate it real quick. And then if that comes back as valid, then great. We upload it into our sequencer.
We get less than a 2 % bounce rate. Everybody's happy. But if you go to multisets and you go to validate, it's like, nope, incorrect.
And you go, okay, cool. I'm going to go to the next data provider. So maybe I go to Blitz.
Maybe I go to AI Arc or whatever. Like you can build your waterfalls from there. And I built these skills so that you can modify and tailor it, no pun intention, but tailor it to your specific workflows.
And so the skill should be able to make it so that you pick a search provider or you do multiple search providers and you can have them go through. all the results then you find it and then it just continues to water through everything and if it's invalid it'll go until it finds a valid one and so when you're done with that you get something like this where we have a full list and this is a specific list where you can see we specifically wanted a company whose head count was over 250 but it had at least 70 people in their sales department So we wanted people that have a certain large company and they had to have a minimum sales department size.
He was able to work through all of that stuff and give me final output files. And then boom, now I have 380 ,000 sales leaders that I can now email as well as the companies that qualify for that. I could take that list of companies and put it into any other search provider like Blitz or Prospeo or, you know, there's millions of them out there.
In fact, I heard lead magic just released a unlimited search API. So I'm going to be testing that out later today. And so once you do that, step four is going to be like your export.
It's going to be like, where do you actually. to send this data do you want to upload it to your sequencer do you just need it in your crm like what's it look like and so it'll make sure to ask you where do you want everything to go when you're all done and so you just tell your agent use this endpoint to upload my leads to xyz campaign and it does it with no technical know -how needed and so for us our export is a little bit more awkward than yours would be because we're in a bit of a unique situation we send our emails through an enterprise partner we call halo i can't name them because we have a very strong nda with them but that's because it's an enterprise grade software they're currently in the beach B2C space.
They want to move into the B2B space, but they can't be publicly known for that yet. So we've coded them in Halo. And so we have to do some super particular stuff like there, like we have to splice our lists into 5 ,000 chunks and we have to upload them manually into the Halo software.
And so on our end, like we have a third skill that is useless to you. So we don't include it here. Um, that actually takes over one of our laptops, clicks through the UI with the browser agent and uploads the files to the sequencer overnight.
But that's not something you need to worry about because if you're sending through something like instantly or smart lead, they have APIs to like make that super easy for you. Eventually we'll have an API on the B2B platform that Halo is making for us, but it just isn't there yet.
But also with this, you know, this isn't just cold email specific, right? The same list, you can do it for LinkedIn outreach, for calls, for LinkedIn ads, where you can often use, you know, that profile URL, just straight out of a search and just skip the waterfall Richmond completely. If you just need a list of LinkedIn profiles to target with LinkedIn ads, for example, you can also modify these skills to like, Hey, I want to do personalization on each.
Cool. You can dispatch an eight, a sub agent inside of the session, and you can have it researched line by line in order to add personal. personalization we do not do personalization anymore because the about six months ago it stopped being effective and now if you personalize there's like no difference between if you personalize an email and if you don't and so then why pay more to get the same outcome and we still test it day to day and we still have seen that personalization is just a total waste of spend makes you feel good makes you feel fancy but at the end of the day dollars talk so for example the campaigns we've ran for rb2b which generated them hundreds of thousand dollars every single month are still some of the best campaigns we've ever ran and there was no personalization in them ever the email was literally here's what the tool does do you want it reply with yes because at the end of the day a good offer is the only thing that matters and believe me we tried there was nothing at all that we could do to further research and personalize an email for somebody that suddenly made them want to buy things more at the end of the day is your product good or not the other thing we don't do is that we don't really score leads before we send to them you know and of course you could do this if you're doing like more targeted stuff you can use this workflow to like the core framework here is here's how to use your local machine to actually be able to process lists for you in a way that like once again i try
to make it so the skills you could take them, you could modify them and put it into your workflow. So it's actually like helpful to you. And so I don't really see the point in scoring.
Our personal take is let's just email a client's total visible market with a single email every 30, 60 days. So we don't burn the market. So we don't spam people too much and just make a bad name for our clients.
And that means just one email, not a four part sequence because we're trying to get them by like right now, right now this week. And of course you like have intent data that somebody is in the market today. Okay.
Maybe that justifies a four part sequence. But once again, it's not exactly what we're doing. We're not taking like an ABM approach.
We just like to monetize the total addressable market and make millions of dollars instead of just trying to generate a really cool, fancy campaign that makes us feel good. And so the last thing we'll leave you with is that this is a very, very small slice of what we do with Fable 5 .6. And so what I did, I started a session up and I was like, hey, can you help me understand everything I've done in the last month with all of my local coding sessions?
And so you can see it on the screen real quick. You can pause to read all of it if you want. But basically, there's a ton of work that I do with these.
In fact, I even created like an AI pool. thing with this thing called a CLI proxy to make it so that I can actually rotate through. I've seven different fable subscriptions right now, and they just all rotate through.
It maximizes the one that's about to expire the soonest. And I have four codex accounts. So I can be able to go through stuff.
I mean, you can see here, you know, I've started up something like, what is it? 6 ,000 agent sessions, just running my company day to day. And so this is just a small preview of all the stuff that we're doing.
And I'm actually working on a scripting on a video right now and putting my thoughts together. I'm like, what the exact system is that I'm using in order to have this type of output and get you know, $28 ,000 of spend.
In fact, let me show you that real quick. Yeah, so here you go. You can see, basically, if I were to pay via the API, I would normally spend $29 ,000 for how much usage I've done in the last 30 days, which is pretty wild when you think about it, because at the end of the day, it's just a bunch of subscriptions.
And as a pro tip, I just found this out the other day, is that you have the $200 plan and the $100 plan with Claude. I thought the $200 plan meant that you get 20x more usage, but that's not true. You only get a 20x multiplier on the five -hour usage, which is pretty crazy.
So I'm actually about to update all of my subscriptions. be $200 each. I'm just going to do the $100 plan because there's not a difference in weekly usage at all.
And by the time I consume one of these with the five hour usage, I'll be able to go to the next one. And so I'm pretty sure I could literally half my cloud bill like tomorrow and still get like the same inference usage out of it, which will be pretty, pretty sick. So I'm gonna start making a few more things about that I'm going to share like the AI pool thing and like what I've been able to do with it and so on.
But like we use agent stuff like crazy. And then also down the road, I'm also working on something right now where I'm actually building a cloud engine again. So that way, my team they can still use their local sessions.
But one of the bottlenecks that they're running into is that if we're processing multiple lists for a client, it's really hard to get those queued up properly. So I'm creating a cloud session. And so they can use their local agents on their computer and they just queue things up in the cloud.
And then it'll process it at that half a million things per second. And even then my agent said that we, if Fable 5 .1 found some efficacies and it believes that we can actually get up to half a million in a second. So who knows, we'll be able to do that.
That'll be pretty cool if we can. So make sure you're following, subscribed or whatever, whenever we talk about this stuff. I just like talking about tech, business, what we're actually doing day -to -day to actually like.
Run a company, like scale it, grow it up, all of that kind of stuff. So make sure you're following me wherever you like to follow me for that. And so to button all of this up, if you ever want us to run cold emails for your company, click the first link in the description and book an intro call with us.
We will map out your total addressable market and show you exactly how we would reach 100 % of it every 30 to 60 days and turn cold email into a money printing machine for you. So again, that's the first link in the description. And the second link in the description should be all the skills that I talked about.
And I hope it serves you well. Lastly, if you want to see the results of this. versus once it's running, watch this video next where I break down how we built the system for Fixer AI.
The Hook
The bait, then the rug-pull.
One teammate moved 8 million cold-email leads through a 16GB MacBook Air, with none of the SaaS stack that used to cost this agency $25,000 a month. What replaced it: two small Claude Code skills, one language, one pure code.
Frameworks
Named ideas worth stealing.
07:29model
The Lead Waterfall
Search
Find
Validate
A three-step cascade: search finds people matching criteria, find pulls their contact details from a data provider, validate confirms the contact is deliverable before it's accepted. An invalid result falls through to the next provider instead of being accepted.
Steal forany outbound list-building workflow, cold email or LinkedIn
03:05concept
Two Tools, One Job
lead-waterfall-processing (language skill)
sequencer-list-scrubber (code skill)
Split an agent workflow into a language-only skill that interviews for intent and produces a plan, and a pure-code skill that executes deterministic transformations like deduplication and formatting.
Steal forany agent workflow currently making an LLM read a spreadsheet row by row
CTA Breakdown
How they asked for the click.
VERBAL ASK
14:35link
“click the top link in the description to book an intro call with us”
spoken direct ask over talking head, paired with a second description link to the two free skills; repeated again in the final 30 seconds alongside a next-video tease
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
A screen-recorded walkthrough plugs Clay's new CLI straight into Claude Code, turning a single spoken prompt into a scraped, verified, cold-email-ready B2B prospect list.
RyMac wires Claude Code straight into Consulti and the ICM folder system, live and unedited, to replace a four-tool cold email stack with one same-day pipeline.
An 18-minute walkthrough of wiring Claude Code into Clay's data platform to source, enrich, and write cold email copy for 50 leads from one natural-language prompt.
A 16-minute walkthrough of replacing a $99+/month ManyChat plan with a free, self-hosted Instagram DM automation, installed end to end by a Claude Code agent skill.
A lightweight alternative to built-in computer-use tools: one markdown skill and a handful of Python scripts that let a coding agent click, type, and screenshot its way through any desktop task.
Anthropic's agent-skills team explains why they stopped building a new agent for every job, and the four habits that make one general-purpose agent actually reliable.
00:14
00:22
00:35
00:45
01:01
01:20
01:40
01:44
01:57
02:09
02:23
02:31
02:42
03:05
03:16
03:32
03:38
03:50
04:15
04:23
04:31
04:44
05:01
05:07
05:19
05:32
05:53
06:04
06:20
06:27
06:38
06:49
07:01
07:12
07:35
07:41
07:54
08:13
08:19
08:30
08:41
08:51
09:04
09:10
09:26
09:49
10:05
10:10
10:28
10:35
10:52
11:07
11:19
11:29
11:41
12:04
12:15
12:26
12:37
12:48
13:00
13:11
13:22
13:33
13:45
13:52
14:02
14:18
14:29
14:52
Video of the Day15:24
59:40
18:02
15:58
13:01
09:47