Ahmed Ezat

CEO of | Leading the AI Agents Revolution. AI agents that work in a real browser, the night shift your team never had.

[Ahmed Ezat @therealezat]

"grow the account" is not a goal you can hand to anyone. not a VA, not an agent.

the checklist we use before an agent gets a goal (save it):

1. one number, one date. "40 booked calls by nov 30", not "more leads"
2. split lag from lead. calls and revenue are lag. the agent can only move lead: profiles read, replies sent, briefs shipped. give it both
3. write the daily run as steps a junior could follow on day one, no guessing
4. hard limits in writing: max sends a day, never delete, never buy, stop and ask when unsure
5. a baseline on day one, read from the page, so "progress" means something
6. a daily report: what was done, the numbers, the next step
7. change one thing a week, never five

the lead/lag split is straight from 4 Disciplines of Execution. it works the same whether the worker is a person or a browser.

this is how goals work on katteb: the agent keeps its notes and picks up tomorrow where it stopped. https://t.co/HFatML0JDT

[Ahmed Ezat @therealezat]

I closed a $1,997 deal from LinkedIn with personalized outreach run by an AI agent https://t.co/w5VwTPjXui

This isn’t a “use my tool” post. I’ll show you exactly how to get the same result, step by step, with Katteb or any AI agent you like.

About 10 days ago I started a LinkedIn outreach campaign. The idea was simple: target SaaS products that had just launched on Product Hunt, every single day.

Here’s what the AI agent did every day:

Opened Product Hunt, pulled the day’s top 20 launches, and grabbed their website links.

Signed up for a free account on each one using its own inbox, so it could verify the account by itself without coming back to me.

Went through each tool’s checkout and upgrade flow and audited it like a CRO expert, looking for bugs, friction and missed conversion opportunities.

When it found a real problem or a big opportunity, it looked up the founder on LinkedIn and sent a connection request. No note.

That came to 5 to 10 founders a day, depending on how many sites actually had issues.

When a founder accepted, it sent them this:

“I was trying your tool and tried to upgrade. I found an issue I’d bet money is costing you thousands in monthly subscriptions. Happy to share it for free if you’re interested.”

The point of that message was to spark curiosity and get a reply, without it feeling like I was selling anything yet.

When they replied: “Can we jump on a 15 min call? I’ll walk you through the issue and suggest a fix.” Plus my calendar link so they could book whatever time suited them.

Out of 41 founders the agent reached over 10 days, 3 booked meetings. On each call I showed them what the agent had found and how I’d fix it.

Then I offered each of them a 3 month CRO engagement: improving their funnel, their upgrade flow, and the triggers that get users to upgrade.

$1,997/month.

1 of the 3 agreed to start with a one month trial after seeing the quick value from our call.

I wanted to share this because it’s pretty much the first high-ticket deal I’ve closed through an AI agent since I launched https://t.co/w5VwTPjXui 2 months ago.

What made this campaign different:

Targeting. I went after founders who were launching. It’s the most important moment for their business, and they’re ready to pay to get seen and get subscribers.

The offer. I didn’t sell in the DM. I offered free help that genuinely makes a difference, because it was customized to them.

And the sale happened face to face, in a live 15 to 30 min meeting.

[Ahmed Ezat @therealezat]

From $440 to $1,000 MRR in just 30 days.

No paid ads.
No cold outreach.
No BS.

Just two growth strategies:

→ Programmatic SEO
→ Dynamic Paywalls

That's a 127% increase in MRR in one month.

Want the exact playbook?
Follow along. I'll break down everything in my next thread.

[Ahmed Ezat @therealezat]

People will prefer to pay more for Claude and keep prompting the AI to do things manually, rather than pay $19 for https://t.co/16DCHVxcuU unlimited usage goals driven AI agents just because they don't know!

[Ahmed Ezat @therealezat]

agency retainers priced by deliverables break the day agents do half the deliverables. the work per client drops, the value to the client doesn't.

what holds up better: price the number the client already tracks (booked calls, leads, reviews answered), plus a floor for the hours you still own.

how are you pricing it now?

[Ahmed Ezat @therealezat]

@MichaelSaruggia the piece that makes the high IQ version work: a suppression sheet. anyone who replied "not now" leaves the rotation and gets their own follow-up date; bounces and unsubscribes leave forever.

skip it and every cycle re-emails people who already told you when to come back.

[Ahmed Ezat @therealezat]

@levikmunneke math holds from month 2. new domains usually need 2-3 weeks of warmup, so month 1 is $307 for close to zero calls.

also 30k emails / 50 inboxes = 600 each, ~30/day, right at the usual per-inbox ceiling. no headroom, so plan spare inboxes for burnout.

[Ahmed Ezat @therealezat]

@Origin_AI_01 small fix to step 1: "sales" is a category, not an outcome. "12 qualified demos booked by Oct 31" is one, and it gives step 4's Review something to check against.

for step 8, sort actions by reversible or not: drafts run free, sends and spend wait for you.

[Ahmed Ezat @therealezat]

@ai_explorer25 context: this is Nadella on BG2 from Dec 2024, and it's aged well.

the practical takeaway for agencies: if the logic leaves the app, each client's rules (approval limits, brand rules, who signs off) need one written doc the agent reads, not a person's head.

[Ahmed Ezat @therealezat]

worth adding to "checks whether links went live": is the link followed, and is the page indexed? many directory profiles are nofollow, and DR76 describes the domain, not the one profile page holding your link.

log rel + index status per link, then drop the sites that give neither.

[Ahmed Ezat @therealezat]

@ycombinator the "harness as intelligence" point transfers to business agents too. before swapping models, change the harness: a written goal with a number and date, a log of what was already tried, a stop rule when it drifts.

do robot evals score harness and model separately?

[Ahmed Ezat @therealezat]

@PrajwalTomar_ the step i'd keep human: launch. have the agent create every campaign as PAUSED (the Marketing API lets you set status on create), then a person flips it live after checking budget, audience and landing URL.

policy checks catch copy. they don't catch $500/day typed as $5,000.

[Ahmed Ezat @therealezat]

@korzhov_dm @bot add a change log between 4 and 7: every bot edit gets a line (what, before → after, date), judged 7 days later.

Meta can send an ad set back into learning after a significant edit, so daily +20% bumps can keep it there. cap: 1 budget move per ad set per week.

[Ahmed Ezat @therealezat]

a task list for an agent: "post 3x, reply to leads, update the sheet." it does all 3 and nothing moves.

a goal: "booked calls 4 → 10/week by Oct 31" (Doerr's from X to Y by when).

the agent picks the tasks. you review the number weekly, not the checklist.

https://t.co/bd384E2fir

[Ahmed Ezat @therealezat]

NN/g's eye-tracking studies back this up (the F-pattern): people read the top lines and the left edge. so put the outcome + a number in the headline, front-load the first 2 words of every line, and show a real product screen above the fold. then watch 10 session recordings and cut what no one reaches

[Ahmed Ezat @therealezat]

@iamAlexTurnbull hi from another founder building AI agents (Katteb). worth posting the monthly version of that goal too: $10M ARR in 36 months is ~$250k net new ARR a month, even counting the $1M you carry over. that makes a slow month visible early. which leading metric will you share weekly?

[Ahmed Ezat @therealezat]

@Fetch_ai i'd add a 7th: can it tell you it's off track? breaking a goal into steps is easy. the hard part is re-reading the number every run ("40 of 300 leads, day 12 of 30") and changing the plan before the deadline, not after. without that it's a task runner with a nicer prompt

[Ahmed Ezat @therealezat]

the "loops in a real person" part is the piece worth copying for any agent. a useful handoff has 3 things: what it already tried, the exact blocker (a phone call, a missing doc, a signature), and the one decision it needs. a handoff that just says "failed" costs you more than doing it yourself

[Ahmed Ezat @therealezat]

the split shows up in where the saved hours go. high-agency people spend them checking output against a written target; everyone else spends them on more prompts. simple rule: every hour of agent work gets 10 minutes of review against a number + date. skip it and you just ship mistakes faster

[Ahmed Ezat @therealezat]

@mhp_guy the 10 auto shops example is a solid agency offer. one step i'd add before the email: have the agent check their FB/IG too. plenty of 300-review shops use a FB page as the site, so the pitch becomes "your FB page can't take bookings", not "you have no site"

[Ahmed Ezat @therealezat]

@lukepierceops #2 and #6 work best as one habit: make the daily update a single line in the client's units. "intake: 38 forms done, avg 4 min (was 45), 0 errors, 1 flagged for you." they can forward that line to their boss, which a Loom never gets. which number do your clients ask about first?