harry jons retweeted
Jev is a textbook example of Clayton Christensen’s Innovator’s Dilemma playing out in AI.
It is commoditizing the bottom of the ML market: classifiers, routers, scoring and context triage. For frontier LLM labs, these workloads are not even worth pursuing. Revenue per decision is tiny, margins are razor-thin, and their entire infrastructure is optimized around selling expensive reasoning and token generation.
They cannot simply put a smaller LLM on fast inference chips like Cerebras and compete. Jev uses fundamentally different infrastructure, built to return typed probabilistic decisions so it’s a pure disruption
harry jons retweeted
FUCK IT. $1.5k/m to $260k for my boy chris
($55k in last 5 days)
here's exactly what we did for him:
- built every video around chris competatiive edge, not what everyone else was doing
- scripted around his buyer's objections so leads showed up pre-sold
- funneled every view straight into his case studies, lead magnets, and offer
- optimised for who was watching, not how many
results:
- 400k+ views
- three video did $100k in sales
- multiple six figures printed from the channel
- got so booked he raised his prices and still kept closing
this is what happens when youtube is built as a sales asset
in short a BUSINESS channel.
(pushing for $100k/m next month)
i’ve already generated $500k+ for b2b & b2c offers
if you want to add an extra $30k–$100k/m to your business
DM me "YOUTUBE" and let's talk
harry jons retweeted
Jensen is right.
Go back and read my lawsuit post from yesterday. Discovery is gonna rip the door wide open on liability
🚨BREAKING: Jensen Huang just EXPOSED Dario and Altman’s "Rogue AI" grift to avoid getting sued into oblivion under EXISTING law
“Don't let this doomsday narrative cause somebody to relieve them of the laws that currently exist.
Go and read between the lines. They're actually not asking for more laws; they're asking to be relieved of the laws we do have.”
ABSOLUTE TRUTH NUKE
harry jons retweeted
A RETARD IN MOTION IS BETTER THAN A GENIUS AT REST
Prediction: @jakepaul becomes a billionaire, buys sports leagues/teams, and is seen as the next @mcuban /@AriEmanuel when he's in his 50s. Underestimating Jake is like thinking a lion cub won't be big enough to kill you in 6 weeks.
harry jons retweeted
never forget that 99.9% of people have no ambition.
I was reflecting about my short stint in the energy space last year. I had dm'ed 100-200 young electrical engineers who worked on the power grid. only 3 replied to me, the rest didn't care to learn more. 1 was the unicorn 1/2
harry jons retweeted
I’m actually stupid to be giving this away for free
this is a cornerstone in our strategy that has generated $40M+ in rev for clients
you can just steal it
harry jons retweeted
Nine times out of ten, traditional backlinks are worthless for local SEO.
However, local backlinks turn map pack scans green.
Here's a list of types of local links we build:
Local chambers of commerce
Neighborhood business associations
Downtown development associations
Local business networking groups
Regional contractor associations
Local homebuilder associations
Independent business alliances
City and county business directories
Local “shop small” directories
Community newcomer guides
Local newspapers
Suburban newspapers
Neighborhood news websites
Local television stations
Local radio stations
City lifestyle magazines
Regional home and design magazines
Local business journals
Community newsletters
Local podcasts
Little League baseball organizations
Youth soccer clubs
Youth football programs
Youth basketball leagues
Youth hockey organizations
Local swim clubs
High school athletic booster clubs
Travel sports teams
Local running and cycling clubs
Minor-league sports teams
Local trade schools
Community colleges
University community programs
High school vocational programs
Apprenticeship programs
Workforce development organizations
Local career fairs
Parent-teacher organizations
School foundations
School robotics teams and band boosters
Local Habitat for Humanity affiliates
Community home-repair nonprofits
Veteran housing organizations
Senior assistance organizations
Food banks and food pantries
Animal shelters and rescues
Homeless shelters
Rotary, Lions, and Kiwanis clubs
Churches and faith-based organizations
Community foundations
Home and garden shows
County fairs
Neighborhood street festivals
Local music festivals
Holiday parades
Charity races and fun runs
Charity golf tournaments
Fundraising galas
Community wild-game dinners
Local sustainability and energy fairs
Independent movie theaters
Community theater companies
Local symphony orchestras
Community arts centers
Local museums
Historical societies
Friends of the Library organizations
Community centers
Makerspaces
Historic preservation organizations
Local real estate agents
Real estate brokerages
Property management companies
Landlord associations
Apartment associations
Homeowners associations
Condominium associations
Neighborhood associations
Local relocation companies
Home inspection companies
General contractors
Custom homebuilders
Kitchen and bathroom remodelers
Roofing companies
Flooring companies
Water and fire restoration companies
Local electricians
Local plumbers
Insulation and energy-audit companies
Architects and interior designers
Local HVAC distributors
Plumbing supply houses
Electrical supply companies
Building-material suppliers
Independent hardware stores
Equipment rental companies
Commercial vehicle dealerships
Vehicle-wrap and sign companies
Local photographers and videographers
Local web designers and IT providers
Economic development organizations
Utility-approved contractor directories
Local energy-efficiency programs
Community weatherization programs
Water-conservation initiatives
Organizations the company already sponsors
Associations the company already belongs to
Nonprofits employees volunteer with
Commercial customers’ project pages
Suppliers’ customer spotlight pages
Local award winner directories
Past speaking engagements and workshop pages
Existing unlinked local business mentions
Existing sponsor pages with unlinked logos
Local websites linking to competitors
harry jons retweeted
why aren't you giving claude code your competitors' best linkedin posts and pulling a lead list out of the likes and comments?
probably the most no brainer thing you can do right now. 4 prompts, about $49 a month in tools.
here's how you get started:
1. pick 3 posts about the problem you solve.
your own first, then fill the open slots from a competitor who writes about the same thing. bigger following, bigger pool, and those people already know what a product like yours does.
2. prompt 1 scrapes everyone who engaged.
claude code starts two apify actors for you, harvestapi/linkedin-post-comments and harvestapi/linkedin-post-reactions, because linkedin keeps reactions and comments as separate lists. cap each run at $2.50 and the pair fits inside apify's free $5. you get one deduped csv with a posts_engaged count per person.
3. prompt 2 researches every lead and tiers it.
origami's agent fills company, role, b2b, employee count, vc-backed, an agency-or-freelancer check and an audience-builder check, each with a one-line reason. run it on 25 rows first and watch the credit balance before you run the rest.
4. the exclusion rule runs before any tier.
out: agency owners, freelancers, consultancies, solo coaches, and anyone commenting to grow their own audience, so creators, ghostwriters and linkedin-growth coaches. the first rule that matches decides the tier.
on screen that came back as nine rows excluded, one tier 3, one still running. three of the excluded were sitting at a fit score of 100. one was a person at frontal, my own company, and the reason read "frontal is a b2b age..."
5. then tier what's left.
tier one is head of sales, vp sales, cro, founder or ceo at a b2b company with 20+ employees. tier two is any gtm role at a vc-backed b2b company. tier three is everyone else who passed rule 1.
6. give each tier its own workflow.
tier one gets email, linkedin and calls from your own team, plus a piece of free work sized to the account. an ae writes and sends every tier two message by hand. tier three runs on an automated email campaign.
7. prompts 3 and 4 build the tier one campaign.
prompt 3 ranks tier one by posts_engaged, so whoever hit more of your posts goes to the top. i asked for 10 profiles and got 4. then a 3-step linkedin sequence: connection request with no note, message 1 after they accept, message 2 six days later to anyone who hasn't replied. about 20 invites a weekday from one account.
544 leads in that table at 2.3 credits each. the tier one campaign runs at 2.4.
full walkthrough in the video below.
harry jons retweeted
We reverse-engineered Instinct's memory system and implemented it with supermemory's architecture
harry jons retweeted
the screen protector is the only thing between him and another 700 dollar sale tbh
harry jons retweeted
Chinese AI labs are scraping the web through Singapore IPs #SEO 👀🙂
harry jons retweeted
Save thousands of dollars a month building software with Grok Build and build better software with an always on software factory.
No more GitHub, lower token costs, lower token usage, and higher quality one shot builds.
Setup:
1. Mac mini or Mac Studio, always on
2. Instruct Grok Build to write a Deployment Manager and build a desktop app.
3. Move all GitHub activity to Grok Deployment Manager
4. Setup Tailscale to run always on Mac from anywhere
5. Create always on skills file that records learnings from every build and include in all build terminal sessions
6. Build and deploy only thru always on Mac or Tailscale session on Mac
harry jons retweeted
There are multiple psychological tricks at play here. Loss aversion, endowment, zero-price, deadline, scarcity, mental accounting and the use it or lose it. Lots of crossover here. Be careful out there.
harry jons retweeted
Here’s an SEO prompt I wish I had years ago:
Stop asking AI:
“Give me low competition keywords.”
Ask it to find weak SERPs instead.
Paste this:
“Analyze this website or niche:
{{URL OR NICHE}}
Find search queries where the current Google results are vulnerable.
Look for SERPs containing:
• Reddit threads
• Quora
• forums
• outdated articles
• thin listicles
• irrelevant results
• pages targeting the keyword indirectly
• small niche sites ranking above major brands
• results with poor search-intent alignment
For every opportunity, tell me:
Target keyword
Search intent
What currently ranks
Why the SERP is weak
What a better page would need to include
Recommended title
Best monetization angle
Do NOT rely on keyword difficulty scores alone.
I want opportunities where the SERP itself shows evidence that Google is struggling to find a great answer.
Return the 20 strongest opportunities.
Then tell me which 5 you would investigate first.”
Keyword difficulty is an estimate.
The SERP is evidence.
harry jons retweeted
here's how to make money with jev on autopilot
it's a super fast decision model, so it picks an answer in a fifth of a second and tells you how sure it is
- put instant pricing on a business that has none
- sell the same thing to bigger businesses for more
- sell a file that decides what an agent may do
- charge monthly to keep all of it running
- costs a fraction of a cent per decision
- runs on hermes or grok bot, on a schedule
just wrote a whole guide on how you can set this up:
harry jons retweeted
Jev is the first time AI feels designed for backends.
Remember the feeling when GPT-3 dropped and you realized the rules of software had changed forever?
I had that exact same feeling messing with Jev yesterday.
For 3 years, we’ve been forcing generative, autoregressive models into places they probably didn’t belong.
The mental model with Jev feels completely different:
LLMs → slow, deliberative, chatty, heavy
Jev → sub-80ms, parallel, calibrated state
Instead of asking an LLM to generate its way through API plumbing, you get something that feels purpose-built for backend systems.
Maybe the interesting shift isn’t making LLMs faster.
It’s building AI that doesn’t need to behave like an LLM in the first place.
harry jons retweeted
Your agent doesn't need a bigger model.
It needs faster decisions.
Jev + Hermes: routing, memory, skill picking, GUI clicks. ~0.4s, fractions of a cent.
github.com/kerpopule/hermes-…
harry jons retweeted
You can create 50 new TikTok & Instagram accounts literally in the next hour with this.
Takes 5 minutes to set up.
1) Sign up on pamba.app and select the free plan
2) Go to the Accounts tab
3) Click "Buy Accounts" and select how many you want
4) Enter your usernames
5) Wait ~30 minutes for the accounts to be created
6) Set up your account warming configuration
7) Start posting your own videos
Not a single person has been banned or bot detected on our platform with over 500M views generated.
Traffic is 95%+ US because all the iPhones are actually in the US and are using US carrier SIM cards, with real phone numbers.
If you need more info please DM me!
I spent $30k on iPhones so that your agents can finally rent them out for going viral on TikTok & Instagram.
No joke, you can turn your Claude/Codex/GrokBot into Mr Beast with this; we've generated over 500M views with this setup.
Here are 5 things you can tell your agents to do today using this MCP server:
1) Create a bunch of TikTok & Instagram accounts (with real US phone numbers & emails)
2) Warm accounts exactly how you like for your niche
3) Capture video ideas from the FYP to understand what's going viral for your competitors
4) Post videos to all of your accounts at once
5) Monitor all of the activity on the phone including screenshots of every action down to the minute
You can literally create 50 TikTok & Instagram accounts in the next 30 minutes and start posting. Nothing gets pre-created or is managed by a human. It's all automated.
We do not share the iPhones with anyone else, they are completely dedicated to you. Every phone has its own unique SIM card, phone number, and never connects to any proxies or wifi.
All of our iPhones are here in San Francisco. Every operation happens through the apps on the phones; we don’t use any external APIs.
Check it out at pamba.app or just give that link to your Claude/Codex.
harry jons retweeted
JEV JUST MADE HALF OF AI ENGINEERING OBSOLETE IN FIVE DAYS.
a model that cannot write one sentence, cannot code, cannot reason - 200x faster, 400x cheaper.
somebody already rebuilt claude code's compaction on top of it and got 4,575 stars in five days. 5 rules before you wire it in.
jev does not generate text. you hand it state and typed questions, it hands back a choice from a set YOU defined, plus a confidence score. it physically cannot return something off-schema. that is the whole product.
-> SEPARATE the deciding from the generating before you touch anything else. most agent loops burn a full generative call on "which tool" and "should i retry" - questions with 3 possible answers. you are paying frontier rates for a multiple choice test.
-> SET your confidence threshold BEFORE you ship, not after the first incident. the calibrated probability is the actual feature here. no threshold means you built a fast oracle and then trusted it exactly as blindly as the slow one.
-> LOG the confidence next to the real outcome, every time. a calibration score that drifts is invisible until you have the paired data. "it was 0.9 confident" means nothing if you never checked what 0.9 cashed out to.
-> TREAT the benchmarks as marketing until someone else runs them. the speed numbers are self-reported and the accuracy reference is an AVERAGE of two frontier models. that is the vendor grading the vendor.
-> STEAL the pattern from fast-jev-compaction, the claude code plugin. it kills the summary step entirely - every tool call and result gets SCORED in one fast request, stale ones dropped or truncated, everything it keeps stays VERBATIM. no summary means no summarization lies. that is the whole shape: decide what survives, do not rewrite it.
third time this year i have ripped a generative call out of a routing step and the loop got faster every time.
would you let something that cannot explain itself decide which half of your context gets thrown away
