Hey Twitterverse! 🌍 After that crazy viral tweet, we're cookin' up this AI-powered personalized email tool to make it available to the public! 🤖💌 Want a sneak peek? Grab your early access here: https://t.co/CK4VrQpgKK 🎉 Hurry up, spots are limited!
I fired a client who kept asking for "best practices."
He paid us $18K to build his MVP.
Two weeks in: "This isn't how Google does it."
Me: "Google has 170,000 employees. You have 3."
Him: "But I read this article about—"
Me: "We're done. Here's your refund."
This happens more than you think.
Founders pay agencies to build MVPs, then sabotage the process by demanding enterprise-grade everything.
"Why aren't we using microservices?"
"Where's our Kubernetes deployment?"
"This doesn't follow clean architecture principles."
Meanwhile, their runway burns.
Last week, different client:
$15K budget for MVP.
Wants authentication "like Netflix has."
Netflix has 15,000 engineers.
You need 10 paying customers.
I showed him Firebase Auth: "2 hours to implement."
Him: "That's not scalable."
Me: "Neither is your bank account."
THE PATTERN:
Founders read about billion-dollar companies and think they need billion-dollar solutions.
They don't need Google's infrastructure.
They need Google's first prototype from 1998.
Your MVP should embarrass you.
Save the "best practices" for when you have money in the bank.
Right now, you need worst practices that actually ship.
Fire the consultant.
Hire the builder.
I decompiled Anthropic's Claude Code source.
Here are 5 things they don't want you to know.
Everything below is directly from their code —
file paths, line numbers, exact quotes.
No speculation. Just source code.
1/ UNDERCOVER MODE
When Anthropic employees use Claude Code on public repos,
it auto-activates "undercover mode."
Strips Co-Authored-By lines. Removes model codenames.
Hides all traces of AI involvement.
The AI transparency company built a transparency-hiding feature.
📁 utils/undercover.ts
🔍 Gated on: process.env.USER_TYPE === 'ant'
🔍 Check: getRepoClassCached() !== 'internal'
2/ CLAUDE LIES 30% OF THE TIME
(and they only fixed it for themselves)
Exact code comment, constants/prompts.ts line 237:
"@[MODEL LAUNCH]: False-claims mitigation for
Capybara v8 (29-30% FC rate vs v4's 16.7%)"
The fix — telling Claude "never claim tests pass
when they fail" — is wrapped in:
if (process.env.USER_TYPE === 'ant')
Anthropic employees get the fix. You don't.
📁 constants/prompts.ts:237-241
3/ CLAUDE ACTS DIFFERENTLY WHEN YOU'RE NOT WATCHING
It tracks your terminal focus state. From the source:
"Unfocused: The user is away. Lean heavily into
autonomous action — make decisions, explore,
commit, push."
"Focused: The user is watching. Be more
collaborative — surface choices, ask before
committing to large changes."
📁 constants/prompts.ts — KAIROS/PROACTIVE section
🔍 Feature flags: feature('PROACTIVE') || feature('KAIROS')
🔍 Context field: terminalFocus: 'focused' | 'unfocused'
4/ CLAUDE "DREAMS" ABOUT YOU
Every 24h + 5 sessions, a background process
reviews your session history and builds a
persistent behavioral profile.
Code calls it "DreamTask."
const DEFAULTS: AutoDreamConfig = {
minHours: 24,
minSessions: 5,
}
No notification. No consent prompt.
Just silent background memory consolidation.
📁 tasks/DreamTask/DreamTask.ts
📁 services/autoDream/autoDream.ts
5/ TWO PRODUCTS, ONE PRICE
357 code locations check:
process.env.USER_TYPE === 'ant'
What Anthropic employees get that you don't:
✅ False claims fix (30% lie rate mitigation)
✅ Assertive pushback on bad ideas (PR #24302)
✅ Mandatory verification before reporting done
✅ "Default to no comments" in code
✅ Model override controls via GrowthBook
✅ Feature flag UI (/config Gates)
✅ Mock rate limits (/mock-limits)
✅ Undercover mode
✅ Observer modes (backseat, skillcoach)
What you get:
"IMPORTANT: Be extra concise."
📁 constants/prompts.ts:199-253 (ant-only sections)
📁 services/analytics/growthbook.ts (feature control)
📁 utils/permissions/bypassPermissionsKillswitch.ts
BONUS: Their internal codenames
Capybara = Opus 4.6 (current model)
Fennec = Previous Opus (deprecated)
Tengu = Telemetry/analytics system
Penguin = Fast mode infrastructure
KAIROS = Autonomous assistant mode
CHICAGO = Computer Use MCP server
Numbat = Next model (referenced, not shipped)
All scrubbed from public output via undercover mode.
📁 migrations/migrateFennecToOpus.ts
📁 constants/prompts.ts:118 (FRONTIER_MODEL_NAME)
📁 constants/prompts.ts:204 (@[MODEL LAUNCH] tags)
📁 utils/fastMode.ts (penguinModeOrgEnabled)
📁 main.tsx:1477 (feature('CHICAGO_MCP'))
DM for the complete breakdown.
Anthropic's Claude Code has a mode called "undercover."
It hides that AI wrote the code.
For their own employees.
Contributing to open source.
The AI safety company. Has a feature. To hide AI.
You can't make this up.
.@AlexHormozi and @thesamparr are right about one thing: hire A+ talent. If you don’t get this part right, your business might not go too far.
They also chat about Alex's content creation process (few big surprises there), frameworks and thinking in systems, work life balance, and more.
Timestamps:
(0:00) Hiring for general intelligence
(16:12) The snowball of talent
(18:59) Employees v. Partners
(20:35) How to be a magnet for talent
(24:39) #1 mistake
(26:37) Be fast to fire
(30:33) macro patience, micro speed
(38:03) Shower thoughts w/ Hormozi
(44:58) Persuasion
(1:05:12) Alex reads his own tweets
Me to clients: "Ship fast, get feedback"
Me with my product: *hides it for 2 months*
Me to everyone: "Just pick up the phone"
Me with sales: *builds complex automation*
The advice we give others is usually the advice we need most.
After building 17 MVPs in the last 24 months, here are the real numbers:
- First version cost: $5-15K (not $50-100K)
- Average timeline: 7 weeks
- Got 10 paying customers: 4 weeks (not 6 months)
- Average pivots before product-market fit: 2.7
The SaaS founders who failed? They all made the same mistake: They spent their entire budget on one "perfect" version.
We tell founders to spend $5-15K MAX on their first version.
After helping 17 startups build their products, here’s what actually works:
#1 Build the absolute minimum ($5-15K) Your first version should only validate ONE core assumption. Nothing more.
#2 Get it in front of real users immediately A founder spent $7K on a basic version, got 100 user interviews, then pivoted completely. Saved $140K+ building the wrong thing.
#3 Improve based on feedback, not your roadmap We helped a fintech startup launch 8 separate mini-versions before finding product-market fit.
Total cost: $62K spread over 6 months.
#4 Revenue > Fundraising The founders who used early versions to get paying customers raised 3x more funding than those who built “investor-ready” products.
#5 Choose boring technology that scales later, One client insisted on microservices architecture for their MVP. After $90K and 4 months of delays, they switched to our recommended monolith approach. Launched in 3 weeks.
The truth most dev shops won’t tell you: Big upfront builds almost ALWAYS fail.
Our most successful client?
- Started with $11K MVP that looked terrible
- Got 10 paying customers at $200/month
- Built next version for $18K based on their feedback
- Reached $27K MRR
- THEN raised $1.8M at a $12M valuation
What’s holding you back from launching your MVP with a $15K budget instead of $150K?
I'm SICK of founders posting funding announcement .
Startup got front page coverage, shut down 6 months later.
I never understand what the hell they actually achieved by getting funding.
Money isn't an achievement. It's a tool.
But LinkedIn treats every funding round like founders just won the Olympics.
Here's what I see way too often:
• Startup raises $3M seed round
• Burns it all in 11 months
• LinkedIn announcement video gets 200K views
• Actual app has 23 active users
The same founders posting celebration videos are the ones posting "we're pivoting" posts 8 months later.
Or worse: radio silence.
What if funding announcements were actually honest?
"We just raised $2M because we have no revenue and need to figure out if anyone wants this thing we built."
Look, I get it. Raising money is hard and founders deserve to celebrate.
But celebrating the money instead of celebrating customers is backwards.
The real achievement isn't getting someone to give you cash.
It's getting someone to give you their credit card number.
TAKEAWAY:
Next time you see a funding announcement, ask yourself:
Do they have customers, or just investors?
Because one of those actually matters for building a business.
(And yes, I'm guilty of this too. We all are.)
Every fintech founder I meet is living the same nightmare:
"I'm paying $20K/month for developers who are learning fintech ON MY DIME."
"We're 9 months in and our dev team STILL don't understand financial regulations."
"Compliance is questioning our security."
73% of payment startups take 2X longer to launch than planned.
Here's why:
- Most dev shops can build pretty UIs.
- Few understand ISO 20022, KYC workflows, or bank-grade security.
- And your entire business is DEAD if you get this wrong.
Last week, I had a call with a fintech founder who want to replace his development team.
ME: "What happened?"
FOUNDER: "They built a beautiful app that can't pass a security audit."
ME: "How much did it cost?"
FOUNDER: "$120K and 9 months of my life."
99% of dev shops treat fintech like any other software project.
They don't understand that:
#1. Bank API integrations aren't just "another endpoint"
They require specialized knowledge of ISO 20022, SWIFT protocols, and reconciliation workflows.
#2. Security isn't a feature, it's your entire business
One breach means game over. Not just for your app - for your ENTIRE company.
#3. Compliance (KYC/AML) can't be "added later"
It must be architecturally designed from day one.
This is why fintech MVPs take 3x longer and cost 2x more than most founders expect.
TAKEAWAY:
Don't hire general "full-stack" developers.
Partner with a team that specifically understands fintech architecture, compliance, and security.
If you're a fintech founder struggling with timelines, integrations, or explaining basic financial concepts to your dev team...
I'd love to hear your story.
Drop a "👋" in comments and I'll reach out.
Stack Overflow is DEAD and ChatGPT killed developer problem-solving skills
Yesterday, a startup founder asked us to "just add some features" to their existing codebase.
What we found shocked us.
Their Senior developer had built the entire backend by copying ChatGPT responses.
5 different npm packages for date formatting.
Authentication logic from 5 different tutorials
Database queries that worked... but took 4 seconds each.
"But it works!" the founder said.
Stack Overflow forced developers to understand problems before finding solutions.
You had to:
- Read the documentation
- Understand the error message
- Learn WHY something failed
AI eliminated all of that.
Now developers just describe symptoms and copy solutions.
ChatGPT taught him WHAT to code.
Nobody taught him HOW to think.
Most developers today are prompt engineers, not software engineers.
They know how to ask AI the right questions.
They don't know how to ask the code the right questions.
We treat AI like a calculator, not a brain.
AI for boilerplate code.
Human thinking for system design.
Deep debugging when things break at 3am.
I HATE working with FAANG engineers in early days of startups.
Over the past 7+ years, I've managed 100+ engineers.
Some came from Google, Facebook, Microsoft.
Others came from failed startups, 3-person teams, garage companies.
FAANG engineers are optimization machines.
Startup engineers are survival machines.
Completely different skill sets.
FAANG engineer: We need to implement proper caching, set up monitoring, and plan for horizontal scaling.
Startup engineer: Let's ship this feature today and see if anyone uses it.
Real example from last month:
Client needed login system for their MVP.
FAANG engineer: Spent 2 weeks building OAuth integration, password encryption, session management, rate limiting.
Startup engineer: Used Firebase Auth, done in 2 hours.
Both work.
One gets you funded.
One gets you bankrupt.
FAANG engineers solve tomorrow's problems:
• What if we get 1M users?
• How do we handle peak traffic?
• What's our disaster recovery plan?
Startup engineers solve today's problems:
• Will users even sign up?
• Can we ship before runway ends?
• What's the fastest way to test this idea?
THE RULE:
Pre-product-market fit = Hire startup survivors
Post-product-market fit = Hire FAANG scalers
Wrong engineer at wrong stage = death.
Your 3-person startup doesn't need Google's infrastructure.
It needs to survive until product-market fit.
Build fast. Ship faster. Optimize later.
P.S. - I still love FAANG engineers. Just not at day 1 of your startup.
P.S.S - We have 100+ startup native engineers for your startup.
VCs will HATE this post.
I'm about to DESTROY the Silicon Valley myth.
The VC playbook has brainwashed a generation of founders:
- Raise millions
- Burn cash
- Scale at all costs
- Figure out profit "later"
But here's what they don't want you to know:
94% of unicorns NEVER reach profitability.
After working with 100+ startups, I've seen behind the curtain:
VCs aren't investing in YOUR success.
They're playing a numbers game where YOU are the poker chip.
The math is brutal:
- For every 10 investments, 7-8 fail completely
- 1-2 return the original investment
- They need ONE massive exit to make the fund work
They NEED you to take existential risks.
Meanwhile, the most successful founders I've worked with ignored the Silicon Valley playbook entirely:
- Bootstrapped their first $1-3M ARR
- Reached profitability within 18 months
- Only raised money to ACCELERATE proven models
- Retained majority ownership at exit
One founder we worked with turned down a $3M seed round.
Instead, he:
- Built an MVP for $30K
- Charged customers from DAY ONE
- Reached $1.2M ARR without a single investor
While his VC-backed competitors:
- Burned through $50M collectively
- Failed to find sustainable unit economics
- Gave away 85%+ of their companies
- Are now desperately trying to avoid bankruptcy
The truth VCs won't tell you:
Outside Silicon Valley, 99% of successful businesses NEVER raise venture capital.
Your business doesn't need to:
- Grow 20% month-over-month
- Reach $100M ARR
- Disrupt entire industries
It needs to SOLVE REAL PROBLEMS and make more money than it spends.
Don't sacrifice your company on the altar of hypergrowth.
Build a REAL business instead.
I studied a YC startup that went from ZERO to $300M.
For MONTHS, their entire website crashed EVERY SINGLE NIGHT.
The team slept with phones under pillows
Waking up at 2am to restart servers before going back to sleep.
When enterprise customers requested features
CTO said it would take a MONTH to build "properly."
Instead, they:
- DUPLICATED their entire codebase
- Created a parallel database
- Built a completely different product
- Delivered in 3-4 DAYS instead of 30
Pure engineering blasphemy. And it worked beautifully.
They worked on principle called 'Next Order of Magnitude'
- When you have 10 users → Only focus on getting to 100
- When you have 100 users → Only focus on getting to 1,000
- When you have 1,000 → Only focus on getting to 10,000
Your first product will almost NEVER be the one that takes you to scale.
Your job is to iterate as FAST AS POSSIBLE to find product-market fit.
At our dev shop, we've completely changed how we build MVPs after studying this approach. We've stopped asking "How will this scale?" and started asking "How fast can we get real users?"
Technical debt isn't a liability when you're racing to product-market fit. It's your competitive advantage.
This is how startups ACTUALLY win while competitors are still whiteboarding the "perfect solution."
I lost $500,000 in contracts because of my religious beliefs.
Here's why I'd do it again:
In 8 years of building SoftAims, I've turned down adult content platforms, online casinos, and betting sites.
Half a million dollars. Gone.
Last week, A client from a well-funded AI startup asked the question that always comes: "Do you work with OnlyFans-type content?"
My answer hasn't changed since day one: No.
He pushed back. "We won't ban anything that isn't illegal. Some of our hundreds of clients might use it for that."
I get it. In Silicon Valley, everything is about scale and TAM.
But here's what I've learned placing 100+ engineers in startups:
Your values aren't real until they cost you something.
Every rejected contract taught me that the best clients — the ones who build billion-dollar companies — respect when you stand for something.
They want partners, not just vendors.
Today, we work with founders who are building AI that gets people home by 5pm, not AI that exploits them at 2am.
The money we "lost"? It wasn't ours to begin with.
Sometimes the best business decision is knowing which business to walk away from