Most people think AI is saturated.
330 million businesses still run on spreadsheets.
Here's what the market actually looks like:
🔘 220M businesses - no automation at all
🔘 90M - basic digital tools, nothing more
🟠 9M - paying for AI services
🔴 1M - actually using AI agents
330 million businesses still run on spreadsheets.
Here's what the market actually looks like:
🔘 220M businesses - no automation at all
🔘 90M - basic digital tools, nothing more
🟠 9M - paying for AI services
🔴 1M - actually using AI agents
That tiny red sliver at the bottom of this chart?
That's your competition.
The gray? That's your client list.
You don't need to build a product. You don't need a massive following. You don't need to be technical.
You just need to know how to walk into a business that's still running on gut instinct and phone calls, and show them what AI can do for them.
Content pipelines. Lead gen systems. Automated workflows. Customer support bots.
Real problems. Real retainers. 330 million businesses that haven't figured it out yet.
The window doesn't stay open forever.
❤4
the most simple Claude Code setup found takes 5 minutes and gets 99% of the job done...
instead of one AI doing everything, you split it into three:
Agent 1, the Architect
> reads your request
> writes a technical brief
> defines scope and constraints
Agent 2, the Builder
> reads the brief
> builds exactly what it says
> nothing more, nothing less
Agent 3, the Reviewer
> compares the output to the brief
> approves or sends it back with specific issues
if rejected... the Builder fixes and resubmits
this loop catches things a single agent would never flag because it can't critique its own decisions (pair it with Codex using GPT-5.4 for best results)
instead of one AI doing everything, you split it into three:
Agent 1, the Architect
> reads your request
> writes a technical brief
> defines scope and constraints
Agent 2, the Builder
> reads the brief
> builds exactly what it says
> nothing more, nothing less
Agent 3, the Reviewer
> compares the output to the brief
> approves or sends it back with specific issues
if rejected... the Builder fixes and resubmits
this loop catches things a single agent would never flag because it can't critique its own decisions (pair it with Codex using GPT-5.4 for best results)
🔥1
the market is expecting 1 million products to launch in the next 24 months...
the competition for building software has never been so hard
here's what i'd suggest: don't build the product
build the infrastructure around the product tsunami:
- distribution tools
- launch platforms
- audience-building services
- analytics for creators
sell shovels during the gold rush is a cliche because it keeps being right
the smartest founders right now aren't building apps... they're building the rails apps launch on
the competition for building software has never been so hard
here's what i'd suggest: don't build the product
build the infrastructure around the product tsunami:
- distribution tools
- launch platforms
- audience-building services
- analytics for creators
sell shovels during the gold rush is a cliche because it keeps being right
the smartest founders right now aren't building apps... they're building the rails apps launch on
🔥1
Most people don’t realize this:
You get 1,500 free daily requests to Gemma 4 31B on @GoogleAIStudio.
That’s plenty of free inference (imo).
And you can route it into @NousResearch Hermes Agent via Vercel’s AI Gateway:
1. Create an API key on Google AI Studio
2. Add it under BYOK (Google) in Vercel AI Gateway
3. Create a Vercel Gateway API key
4. In Hermes → select “Vercel AI Gateway” + your Google model
Now all your Google model requests route through your free AI Studio quota.
Basically: free 31B model access inside your agent stack.
(Tradeoff: not as private as running locally)
You get 1,500 free daily requests to Gemma 4 31B on @GoogleAIStudio.
That’s plenty of free inference (imo).
And you can route it into @NousResearch Hermes Agent via Vercel’s AI Gateway:
1. Create an API key on Google AI Studio
2. Add it under BYOK (Google) in Vercel AI Gateway
3. Create a Vercel Gateway API key
4. In Hermes → select “Vercel AI Gateway” + your Google model
Now all your Google model requests route through your free AI Studio quota.
Basically: free 31B model access inside your agent stack.
(Tradeoff: not as private as running locally)
🔥1
building an AI agent that talks to your business data is probably the highest-leverage thing you can build right now.
everyone's building chatbots or content tools but the real play imo is building an agent that sits on top of your ad performance.. inventory levels or your CRM
anyone can build a wrapper on top of GPT but nobody else has your data, ad spend patterns, seasonal cycles, or customer behavior.
an agent that understands your specific business context gives you an unfair advantage.
i'd recommend setting this up with a hermes agent.
here's a quick step-by-step:
> start by identifying your highest-value data source.
for most people that's their ad accounts, their analytics, or their inventory system. pick one. don't try to connect everything at once.
> connect it to hermes agent.
API connections, CSV uploads, database access. hermes ingests the data and structures it so you can have a conversation with it.
> give it context and instructions.
tell it what matters to your business. what metrics you care about. what thresholds should trigger an alert. what "good" and "bad" look like for your specific situation.
> set up a slack bot and connect hermes to it.
now your whole team can ask questions in plain language right inside slack. "which ad creative is fatiguing this week?" "where should i shift budget tomorrow?" "what's trending toward stockout in the next 7 days?" answers in seconds from your actual data. everyone on the team gets visibility without pulling a single report.
> build the feedback loop.
every time you correct it, every time you tell it "that metric doesn't matter as much" or "flag this earlier next time," it learns. the agent improves its recommendations based on how you actually run your business.
after 30 days the agent knows your seasonality, your cycles, your anomalies..
after 90 days it's catching patterns you wouldn't spot manually and making recommendations you'd pay a consultant $500/hour for. and it just keeps getting sharper the longer it runs.
some examples of what this looks like in practice:
> ecom brand asking "which creative is fatiguing and where should i reallocate budget" and getting an answer in 10 seconds instead of pulling reports for an hour
> DTC founder getting a slack alert when a SKU is trending toward stockout 5 days before it happens
> sales team asking "which leads haven't been touched in 48 hours" and getting a prioritized list with context on why each one matters
> founder asking "what's my burn rate this month vs last 3 months" and getting a breakdown with the anomalies highlighted
every one of these is an agent sitting on top of data the business already has. the data was always there. the interface to talk to it wasn't.
pick your highest-value data source. connect it. start asking questions.
the compounding starts the day you begin. every day you wait is context the agent doesn't have. Source
everyone's building chatbots or content tools but the real play imo is building an agent that sits on top of your ad performance.. inventory levels or your CRM
anyone can build a wrapper on top of GPT but nobody else has your data, ad spend patterns, seasonal cycles, or customer behavior.
an agent that understands your specific business context gives you an unfair advantage.
i'd recommend setting this up with a hermes agent.
here's a quick step-by-step:
> start by identifying your highest-value data source.
for most people that's their ad accounts, their analytics, or their inventory system. pick one. don't try to connect everything at once.
> connect it to hermes agent.
API connections, CSV uploads, database access. hermes ingests the data and structures it so you can have a conversation with it.
> give it context and instructions.
tell it what matters to your business. what metrics you care about. what thresholds should trigger an alert. what "good" and "bad" look like for your specific situation.
> set up a slack bot and connect hermes to it.
now your whole team can ask questions in plain language right inside slack. "which ad creative is fatiguing this week?" "where should i shift budget tomorrow?" "what's trending toward stockout in the next 7 days?" answers in seconds from your actual data. everyone on the team gets visibility without pulling a single report.
> build the feedback loop.
every time you correct it, every time you tell it "that metric doesn't matter as much" or "flag this earlier next time," it learns. the agent improves its recommendations based on how you actually run your business.
after 30 days the agent knows your seasonality, your cycles, your anomalies..
after 90 days it's catching patterns you wouldn't spot manually and making recommendations you'd pay a consultant $500/hour for. and it just keeps getting sharper the longer it runs.
some examples of what this looks like in practice:
> ecom brand asking "which creative is fatiguing and where should i reallocate budget" and getting an answer in 10 seconds instead of pulling reports for an hour
> DTC founder getting a slack alert when a SKU is trending toward stockout 5 days before it happens
> sales team asking "which leads haven't been touched in 48 hours" and getting a prioritized list with context on why each one matters
> founder asking "what's my burn rate this month vs last 3 months" and getting a breakdown with the anomalies highlighted
every one of these is an agent sitting on top of data the business already has. the data was always there. the interface to talk to it wasn't.
pick your highest-value data source. connect it. start asking questions.
the compounding starts the day you begin. every day you wait is context the agent doesn't have. Source
❤2🔥1
Vibe coders are getting sued.
People are launching apps with real users but skipping the boring stuff that can actually kill the product.
A developer with 20+ years of experience just shared the pre-launch checklist every AI builder should run:
→ privacy policy if you collect user data
→ know where user data is stored
→ check security headers
→ scan against OWASP basics
→ look for SQL injection / XSS / auth issues
→ make sure .env values are not leaking
→ check API responses for sensitive data
→ remove secrets from logs
→ never expose API keys in frontend code
→ move keys server-side or behind a proxy
→ add rate limits before someone burns your API bill
This is what most vibe coders are missing.
AI can help you build the app.
But if you launch without security, privacy, and abuse checks...
you didn't ship a product.
you shipped a liability.
People are launching apps with real users but skipping the boring stuff that can actually kill the product.
A developer with 20+ years of experience just shared the pre-launch checklist every AI builder should run:
→ privacy policy if you collect user data
→ know where user data is stored
→ check security headers
→ scan against OWASP basics
→ look for SQL injection / XSS / auth issues
→ make sure .env values are not leaking
→ check API responses for sensitive data
→ remove secrets from logs
→ never expose API keys in frontend code
→ move keys server-side or behind a proxy
→ add rate limits before someone burns your API bill
This is what most vibe coders are missing.
AI can help you build the app.
But if you launch without security, privacy, and abuse checks...
you didn't ship a product.
you shipped a liability.
❤2🔥1
how to get your startup mentioned by chatgpt, perplexity, and claude:
- publish articles that answer the questions your icp asks llms
- make sure your content is well organized and easy to parse
- add faqs with direct answers to common icp questions
- build comparison pages and alternative pages
- fix your technical issues
- post on reddit threads that rank on page 1 of google
- track which competitors get mentioned and which sources llms cite
- monitor your llm mentions and find gaps
okara ai cmo does all of this automatically
- publish articles that answer the questions your icp asks llms
- make sure your content is well organized and easy to parse
- add faqs with direct answers to common icp questions
- build comparison pages and alternative pages
- fix your technical issues
- post on reddit threads that rank on page 1 of google
- track which competitors get mentioned and which sources llms cite
- monitor your llm mentions and find gaps
okara ai cmo does all of this automatically
❤1🔥1
if I had one weekend to master Claude
i'd start here:
- claude 101 : lnkd.in/g75wfW-A
- claude Code : lnkd.in/gsRhsFxT
- stop prompting: lnkd.in/dJgsf93y
- claude Cowork + Obsidian: lnkd.in/dJgsf93y
- claude for your team: lnkd.in/gafjDp35
- no prompt saves you: lnkd.in/erHRYTmt
- claude to sound like you: lnkd.in/eF56s4i8
- set up Claude Cowork: lnkd.in/dxQ87cvk
tldr: substack + reddit + LinkedIn + claude ≈ infinity stone
i'd start here:
- claude 101 : lnkd.in/g75wfW-A
- claude Code : lnkd.in/gsRhsFxT
- stop prompting: lnkd.in/dJgsf93y
- claude Cowork + Obsidian: lnkd.in/dJgsf93y
- claude for your team: lnkd.in/gafjDp35
- no prompt saves you: lnkd.in/erHRYTmt
- claude to sound like you: lnkd.in/eF56s4i8
- set up Claude Cowork: lnkd.in/dxQ87cvk
tldr: substack + reddit + LinkedIn + claude ≈ infinity stone
🔥3