Carolina Posma

carolinaposma.com

I build AI agents that live inside WhatsApp, Instagram, Slack & your website.

Your inbox is full. Your leads aren’t followed up. Your team is drowning in repetitive questions.

You could hire more people.
Or you could automate all of this with AI agents that work inside the apps you already use.

💬 WhatsApp AI Agent – Follows up on leads, books calls, supports customers
📩 Instagram AI Agent – Handles DMs, sets appointments, closes sales
💻 Website Chat AI Agent – Answers questions, boosts conversions
🧑‍💼 Slack AI Agent – Automates workflows & integrates with your tools

No gimmicks. No clunky bots. Just AI agents that feel human and get sh*t done.

Want an AI agent for WhatsApp, Slack, or your website? Let’s talk.
DM me or book a call: www.carolinaposma.com

AI Agents for Whatsapp, Slack & website: founder, implementation & training

FlowGent AI

Implementation Partner & Advisor

Digital Qube Marketing

AI Trainer

ColdAgency

• Led automations for cold email and LinkedIn lead generation for corporate trainers, coaching companies, B2B agencies, recruitment & staffing companies, startups and software companies.
• Developed sales and CRM process automations.
• Provided ghostwriting services for LinkedIn content.

(AI) Automation, Growth & Customer Success Expert

Content Jungle

• SEO & Content Agency
• Shifted focus to automations & AI

Founder

Anywyse

•  Audio learning platform for higher education.

Co-Founder (Operations, Product & Marketing)

I build AI Agents for WhatsApp, Slack and your website for magical conversational experiences ↴

Your inbox is full. Your leads aren’t followed up. Your team is drowning in repetitive questions.

You could hire more people.
Or you could automate 𝘢𝘭𝘭 of this with AI agents that work 𝘪𝘯𝘴𝘪𝘥𝘦 the apps you already use.

💬 WhatsApp AI Agent – Follows up on leads, books calls, supports customers
📩 Instagram AI Agent – Handles DMs, sets appointments, closes sales
💻 Website Chat AI Agent – Answers questions, boosts conversions
🧑‍💼 Slack AI Agent – Automates workflows & integrates with your tools

No gimmicks. No clunky bots. Just AI agents that feel human and get sh*t done.

I’ve helped businesses implement AI to:
→ Convert more leads without hiring more sales reps
→ Save hours on customer support
→ Keep clients engaged 24/7 with instant responses

& I also write a weekly newsletter about AI agents & share free templates and use cases.

Want an AI agent for WhatsApp, Slack, or your website? Let’s talk.
DM me or book a call: www.carolinaposma.com

🚀 Boost your sales and credibility with LinkedIn & blog thought leadership content

Our goal is to help business owners like you get noticed by your potential customers through powerful, engaging LinkedIn and blog content.

👉 Our 100% done-for-you solution will take your LinkedIn presence to the next level and transform you into a LinkedIn thought leader by repurposing your existing website content into engaging posts.

👉 And if you don't have existing content yet, don't worry! Our team of knowledge junkies are here to help you get started by writing high quality and in-depth blog posts that will drive traffic and establish your brand.

It’s time to turn your expertise into a sales driver and establish yourself as a B2B thought leader.

Holy sh*t. Did we really create this carousel with one prompt???

Guys I'm so so so excited. We're currently testing Postiv and the results are blowing our minds

With one prompt we created this amazing LinkedIn carousel

What you're seeing in the video is JUST the first draft of the carousel (you can make any changes) and it's already brilliant

In case you missed it:

With Postiv we're fighting against all the generic AI slop that's being put out here on LinkedIn

You already have years of expertise sitting in presentations, blog posts, and conversations.

The Postiv AI agent turns that knowledge goldmine into high-value LinkedIn posts and carousels that reflect *your original expertise and knowledge*

We're building the best AI agent for LinkedIn content creation with your feedback:

Let me know in the comments what you think or if you want early access

AI Agent & Automation Expert | Building Postiv.ai (AI Agent for LinkedIn Content Creation) | Founder of AI Agents Made EASY | Helping Businesses Scale with Smart AI Solutions

BREAKING: Make and n8n both changed their pricing

Two major automation platforms just changed their pricing. Here's what you need to know:

𝗠𝗮𝗸𝗲.𝗰𝗼𝗺 (𝗔𝘂𝗴𝘂𝘀𝘁 𝟮𝟳, 𝟮𝟬𝟮𝟱)
→ Operations become "credits" 
→ 1 operation = 1 credit (automatic conversion)
→ Your current price stays the same
→ AI modules cost more credits if you use Make’s built-in provider
→ Bring your own API key? Then it will always be 1 credit (= 1 operation)

What this means in simple terms: if you use the native Make AI provider, you will pay more credits. If you bring your own API key, nothing changes.

𝗻𝟴𝗻 (𝗟𝗶𝘃𝗲 𝗻𝗼𝘄)
→  All plans now offer unlimited workflows, steps, and users.
→  A new self-hosted Business Plan includes Git workflows, SSO, scaling, and more (starts at $800/month)
 → Business (self-hosted) and Enterprise plans now charge per workflow execution
 → Community Edition remains free

What this means in simple terms: 
Small users = still free. Larger self-hosted companies = much higher costs.

There’s a lot of backlash in the n8n community as  high-volume self-hosted users report license costs exceeding $10K/month.


𝗪𝗵𝗮𝘁 𝘁𝗵𝗶𝘀 𝗺𝗲𝗮𝗻𝘀:

Make users: Minimal immediate impact, unless using Make.com's built in AI modules. (But I'm curious what the next steps will be for this credit system).

n8n users: Free tier unchanged. Self-hosted business feels overpriced according to users.

𝗧𝗵𝗲 𝗯𝗶𝗴 𝗽𝗶𝗰𝘁𝘂𝗿𝗲:
The automation industry is maturing. Pricing models are evolving to match AI-driven capabilities.

Neither change breaks existing workflows, but both signal where the industry is heading.


Repost to help others understand the pricing changes ♻️

No alternative text description for this image

Everyone talks about LinkedIn growth hacks, but they're missing the easiest one.

You're sitting on a goldmine and don't even know it.

Your LinkedIn data. 

Here's how to unlock the goldmine in 10 minutes:

𝗦𝘁𝗲𝗽 𝟭: 𝗗𝗼𝘄𝗻𝗹𝗼𝗮𝗱 𝗬𝗼𝘂𝗿 𝗗𝗮𝘁𝗮
→ Settings & Privacy  
→ Get a copy of your data
→ Select "Download Larger Data Archive"
→ You will get a list of your connections within 10 minutes, and a csv file with all your posts within 72 hours

𝗦𝘁𝗲𝗽 𝟮: 𝗠𝗶𝗻𝗲 𝗬𝗼𝘂𝗿 𝗖𝗼𝗻𝗻𝗲𝗰𝘁𝗶𝗼𝗻𝘀
→ Export 1st connections to CSV
→ Upload to Clay for prospect research  
→ Identify ideal customers
→ Start targeted outreach

𝗦𝘁𝗲𝗽 𝟯: 𝗧𝗿𝗮𝗶𝗻 𝗬𝗼𝘂𝗿 𝗔𝗜
→ Feed old posts ("shares.csv" in your data archive) to Claude Projects, your custom GPT, or as a knowledge base to your content automation system
→ *Never run out of content ideas*

𝗦𝘁𝗲𝗽 𝟰: 𝗙𝗶𝗻𝗱 𝗪𝗶𝗻𝗻𝗶𝗻𝗴 𝗣𝗮𝘁𝘁𝗲𝗿𝗻𝘀  
→ Analyze top-performing posts
→ Identify best topics & formats
→ Double down on what works

𝗪𝗵𝗮𝘁 𝘁𝗵𝗶𝘀 𝘂𝗻𝗹𝗼𝗰𝗸𝘀:
↴ Warm prospect database from existing network
↴ AI assistant trained on your voice  
↴ Data-driven content strategy
↴ Higher engagement rates

Your growth shouldn't be guesswork.

Your data shows you exactly what works.

Most creators ignore this goldmine.

Don't be most creators.

Repost this to share this growth hack with others as well ♻️

10 months ago I got 3 likes on a LinkedIn post. Yesterday I hit 2 million total views.

What changed? I started showing up consistently, sharing my expertise and bringing value.

But here's what nearly killed my consistency along the way:

I was spending hours every week trying to come up with new angles, writing posts from scratch, desperately trying to turn my knowledge into engaging LinkedIn content. It was exhausting.

So naturally, I tried every AI content tool I could find. They seemed promising at first - but they all had the same fatal flaw.

You input a topic. It spits out templated fluff.

No personality, real insights, or real value. Just generic AI slop that sounds like everyone else's posts.

These tools don't know your expertise. They don't know about that presentation you gave last year that had clients nodding along. Or the client call where you explained something perfectly and they finally "got it." Or the blog post that actually helped people solve real problems.

So we built something different. We realized everyone has years of expertise sitting in old presentations, blog posts, voice notes, and recorded conversations. Gold mines of knowledge just collecting digital dust.

The Postiv AI agent learns from your existing knowledge:
→ Upload your presentations and speaking materials
→ Feed it your best-performing posts and content
→ Add transcripts from client calls and explanations
→ Include case studies, blogs, Youtube videos and anything you've created

Then ask it to create LinkedIn content. The difference is shocking.

Instead of generic AI slop, you get posts that sound like you. Because they're based on your real expertise, your actual insights, your unique perspective on problems you've solved.

We're launching next week and the results are making me excited to post again. LinkedIn is drowning in identical AI-generated content, but you have real knowledge. Real insights. Real value to share.

Comment POSTIV to get early access to the launch deal - I'll DM you.

This is the LinkedIn content tool I've been wanting to build for years. It's finally here.

Building Postiv.ai (AI Agent for LinkedIn Content Creation) | Founder of AI Agents Made EASY | AI Agent & Automation Expert | Helping Businesses Scale with Content & Growth AI Solutions

I stopped typing 90% of my content last month.

And my productivity doubled.

The secret? Voice-to-text AI.

One of the big advantages of AI has of course been voice-to-text recognition. This isn't new.

But after I started using it a lot for my content creation, I noticed something.
 
I discovered that speaking to AI unlocks something typing can't:

☑ Richer context through natural rambling 
☑ Faster idea generation (no typing bottleneck) 
☑ Better explanations (you talk like you think) 
☑ More authentic tone in final output

Here's what I learned:

When I type, I feel lazy and impatient.
When I speak, I give raw, unfiltered context.

That extra context leads to dramatically better results.

There are manyyy use cases for this of course, but 
The workflow:
↴ Speak your rough ideas out loud:
→ Directly into my content creation system with Wispr Flow
→ On the go to your WhatsApp AI agent or Slack AI agent (I like to drop voice notes to my ai agents)

Next:
↴ Let AI capture the rambling thoughts 
↴ Your LinkedIn content automation workflow drafts posts
↴ You make final edits (minimal)

Result: 50% faster content creation with better quality.

(for my content creation system I use Airtable, but it also works in Notion, ClickUp or any other tool that you use to store your posts)

If you haven't tried voice-first AI workflows yet, give it 2 days of experimenting.

You'll never go back to typing everything.

P.S. This post was dictated, then refined. Took 3 minutes instead of 10.

Repost to inspire others with smart automation systems ♻️

Time for a proper reintroduction. 🙋‍♀️

Most of you know me from AI Agents Made Easy → building conversational AI agents for businesses and educating non-tech business owners on how to leverage AI for their business.

But here's the full story of how I got here:

Originally from the Netherlands, I've lived in Lisbon for a couple of years before recently splitting my time between Dubai and Bali (escaping the Dubai heat). When I'm not working on AI implementations, you'll find me in the gym (joined my first Hyrox event this year!) or buried in a fantasy book.

My professional journey took me through:
→ Co-founding my first startup (edtech) right out of uni - where we turned study material into short educational podcasts
→ Running my SEO agency with a team of 6 amazing writers (with LinkedIn content being a huge lead gen channel)
→ Content creation as a LinkedIn ghost writer (learning what actually works on LinkedIn)
→ Lead generation at ColdAgency (discovering that clients with consistent LinkedIn content got 3x better outreach results)
→ AI automation & agent implementation (AI marketing and sales automations such as content creation, WhatsApp AI agents, Instagram DM AI agents, and customer support automation)

And throughout this journey there was a thing that I noticed...

Every B2B business owner I worked with had the same problem: 

They knew LinkedIn content was crucial for getting more customers, but creating it consistently was their biggest bottleneck.

They kept staring at blank pages.

While they all have websites, youtube videos, presentations, blog posts, and webinars full of expertise.

Each piece contains 10-20 LinkedIn posts worth of insights, and each post can get you new clients, but manually extracting them and turning them into LinkedIn posts takes hours.

So that's why I'm now launching Postiv AI → your AI agent team that learns from you and turns your existing expertise into consistent LinkedIn posts and carousels.

My current products & services:
☑ Postiv AI - LinkedIn content tool (launching next week!)
☑ AI Agents Made EASY - Practical AI education
☑ Custom AI implementation partnerships (from AI audits to sales/marketing AI systems and Instagram AI Appointment Setters)

The connection? I help businesses scale with smart AI solutions.

If you're tired of generic AI content advice and want to leverage AI to amplify your existing expertise, let's connect. Feel free to send me a DM! 🤩

P.S. Make sure to sign up to the Postiv waitlist to get the launch deal (link in first comments)

Most professionals still think AI agents are this mysterious, complex technology.

But the concept is actually super simple when you break it down.

So I made this quick explainer to show you exactly what an AI agent is (and how it's different from just using ChatGPT).

Here's what you need to know:

Repost to help others understand AI agents ♻️

What exactly is an AI agent

The hidden cost of manual LinkedIn posting that nobody talks about.

Your brain capacity.

I used to spend 2 hours crafting one post, wrestling with blank page syndrome.

My mental energy was drained before I even started my real work.

Here's the breakthrough:

You don't need more creativity. You need better systems.

The content multiplication hack nobody uses:

Step 1: Audit your existing content goldmine
→ Blog posts gathering digital dust
→ Presentations from last quarter  
→ Client case studies never shared
→ Random voice notes with breakthrough insights

Step 2: Apply the 3-category extraction method
→ Viral content: surprising stats, contrarian opinions, hot takes
→ Educational content: step-by-step processes, frameworks, common mistakes  
→ Authority content: client results, before/after transformations, proven methodologies

Step 3: Use the multiplication math
→ 45-minute webinar = 5-8 viral posts + 6-10 educational insights + 3-5 authority examples = 18 posts
→ One client presentation = 15 posts
→ One newsletter = 8 posts
→ One strategy document = 12 posts

Step 4: Set up in Postiv
→ Upload your existing content goldmine in Postiv .ai
→ Ask Postiv to turn it into engaging content and carousels (yes - this is your vibe carousel designer!)
→ The Postiv AI agent will research based on your knowledge, you just have to do the final edits
→ Publish your posts

I went from 2 hours per post to 10 minutes per post. (So I can instead enjoy a drink on the terrace. ☀️)

Same engagement. Better results. Zero brain fog.

The secret? Stop creating from scratch.

Start repurposing strategically with systems.

Your expertise already exists. You just need to unlock it systematically.

What existing content could you turn into 10 LinkedIn posts this week?

♻️ Repost this to inspire others to start posting on LinkedIn as well!

This Caribbean island is getting rich from AI

(and it's not for reasons you might expect)

Meet Anguilla:
→ 16,000 people
→ 35 square miles
→ Hundreds of millions in AI revenue

𝐇𝐨𝐰?
The world's luckiest government decision.

Back in 1990, internet officials needed country codes.

Anguilla: "Sure, we'll take .ai"

Nobody cared. It was just paperwork.

Today? Every AI company wants .ai as a domain extension.

33 years later, that simple decision prints money.

This post is a simple reminder that decisions that might seem insignificant today may change your future in unexpected ways.

You just need to make that decision, take that action, keep moving and keep trying.

♻️ Repost this to remind others to take action, how small or insignificant it may seem today.

Why most business owners stop posting on LinkedIn after 3 months

Most people create content randomly.

They wake up, think "I should post something," then scramble for an idea.

This approach kills momentum fast.

Here's what I learned after scaling to 20,000 followers in 10 months:

Consistent creators think in 3 strategic categories:

→ Viral content (gets eyeballs)
Tool lists, contrarian opinions, surprising stats, hot takes

→ Educational content (builds thought leadership)  
Step-by-step processes, frameworks, common mistakes

→ Authority content (converts followers to clients)
Client results, case studies, your proven methodologies

Random creators run out of ideas in week 2.
Strategic creators have endless content because they follow a system.

Random creators post when they feel like it.
Strategic creators post consistently because they planned ahead.

The result? Strategic creators build sustainable thought leadership.

Random creators fade away.

Your content doesn't need to be perfect.
It needs to be strategic.

Which categories have you been focusing on in your content?

Why most business owners stop posting on LinkedIn after 3 months

My yesterday use case for Nano-Banana: turning a casual pic into your typical LinkedIn laptop pic 🤫 👩‍💻

No idea what nano banana is? It's the new image tool from Google.

It's insanely good which is why everyone is hyped about it.

You can access it in Googles AI studio: aistudio .google .com

PS. use Postiv .ai to create expert-led linkedin content - it's for businesses who know how much $$$ linkedin provides as a marketing channel when you share valuable expertise-based content

Anthropic shared their complete framework to build AI agents

Here's what you should know:

𝗪𝗵𝗲𝗻 𝘁𝗼 𝗕𝘂𝗶𝗹𝗱 𝗔𝗴𝗲𝗻𝘁𝘀:

→ Task is ambiguous and complex
→ High value justifies token costs  
→ No critical capability bottlenecks
→ Errors are catchable and low-stakes

If you can map every decision path, use workflows instead.

𝗧𝗵𝗲 𝟯-𝗖𝗼𝗺𝗽𝗼𝗻𝗲𝗻𝘁 𝗔𝗴𝗲𝗻𝘁:

𝟭. Environment - Where the agent operates
𝟮. Tools - How it takes actions  
𝟯. System Prompt - Goals and constraints

That's it. Anthropic uses this same backbone for every agent.

𝗧𝗵𝗶𝗻𝗸 𝗟𝗶𝗸𝗲 𝗬𝗼𝘂𝗿 𝗔𝗴𝗲𝗻𝘁:

Your agent only knows what's in its 10K-20K token context window.

When it makes weird decisions, ask: "What context was missing?"

Try completing tasks with only the info your agent has. You'll quickly see the gaps.

𝗢𝗽𝘁𝗶𝗺𝗶𝘇𝗮𝘁𝗶𝗼𝗻 𝗦𝘁𝗿𝗮𝘁𝗲𝗴𝘆:

→ Get behavior right first
→ Parallelize tool calls for speed
→ Cache trajectories to save tokens
→ Show progress to build trust

Simple beats complex. Every time.

Keep iterating on those 3 core components before adding complexity.

Repost this to help others understand AI agent frameworks as well ♻️

Your LinkedIn strategy is backwards if you're getting lots of engagement but no inquiries.

Here's why:

Business owners who monetize their LinkedIn accounts the best understand the difference between HOW and WHAT content.

Here's the difference:

HOW content reveals your methodology:
→ "Here's my 5-step process for..." 
→ "The exact framework I use to..." 
→ "Step-by-step breakdown of how I..."

WHAT content showcases your outcomes: 
→ "This strategy doubled our client retention" 
→ "Why most agencies fail at scaling" 
→ "The insight that changed everything for us"

HOW content attracts people who want to learn your process.

WHAT content attracts people who want to hire your expertise.

The catch? Your content strategy should match your service type.

If you sell courses or training 
→ Share more HOW

If you sell done-for-you services 
→ Share more WHAT

If you consult on strategy 
→ Balance both strategically

Your LinkedIn content should create demand for your expertise, not eliminate the need for it.

What type of content are you sharing more of - and does it match what you actually sell?

Palantir... people either love them or hate them. And they just dropped their AI manifesto: "Working Intelligence"

Their main point:

"You’re being sold an AI future where you’re obsolete or irrelevant. We see it differently."

According to Palantir:
→ Both AI doomers and pacifiers are wrong
→ AI should make workers "more of themselves" not obsolete

What's your take on this?

Employee-led marketing is becoming the secret weapon for LinkedIn growth. And some companies are absolutely crushing it.

Lemlist pioneered this approach years ago:

→ Guillaume built massive personal authority in outreach
→ Roxana became the go-to voice for marketing strategy  
→ Kévin (aka the Cheat-Sheet guy) turned complex tactics into viral content

The employees of Lemlist became industry thought leaders who happened to work at Lemlist.

Now ColdIQ is taking this playbook and supercharging it under guidance of Millie (make sure to give the team of ColdIQ a follow for amazing AI-powered lead gen content!)

Their team is ruling LinkedIn with AI-powered lead generation strategies. Each employee is a mini-brand building authority in their niche.

Here's why this works better than traditional marketing:

☑ Personal brands get 5x more engagement than company pages
☑ People trust individuals more than corporate accounts
☑ Your team becomes a distribution network all with their OWN audience
☑ Authentic content converts better than polished ads

The companies winning on LinkedIn are the ones who've figured out how to activate their entire team as content creators.

Your employees have knowledge. Your employees have networks. Your employees have stories.

Which other teams should we follow on LinkedIn? Tag them below in the comments 👇

P.S. With Postiv your employees can easily create expertise-led LinkedIn posts that are based on your shared company knowledge base.

Employee Led Marketing

𝗖𝗵𝗮𝘁𝗚𝗣𝗧-𝟱 𝗵𝗮𝘀 𝟰 𝗺𝗼𝗱𝗲𝘀. 𝗠𝗼𝘀𝘁 𝗽𝗲𝗼𝗽𝗹𝗲 𝗼𝗻𝗹𝘆 𝘂𝘀𝗲 𝗼𝗻𝗲.

The breakdown:

Auto:
→ Decides fast vs deep based on your prompt
→ Perfect for: general writing, basic coding, research
→ When to use: You don't want to think about which mode

Fast: 
→ Speed over depth every time
→ Perfect for: customer support, meeting summaries, quick checks
→ When to use: Latency matters more than perfection

Thinking:
→ Maximum reasoning power, slower output
→ Perfect for: strategy work, financial modeling, complex debugging
→ When to use: Deep analysis can't have gaps

Pro:
→ Reserved for Pro/Team subscribers only
→ Perfect for: research-grade work, product design, critical decisions
→ When to use: Stakes are too high for errors

Test the different modes. Play with it. Figure out what works best for each of your tasks.

Repost this so others will learn about the different GPT-5 modes as well ♻️

EVERYONE can build an AI agent. Even if you are non-technical.

Most people think you need to be a tech bro and understand complex tools like n8n or code.

Wrong.

I've helped dozens of non-tech founders build their first AI agent in hours, not days.

Here's the simple path they all follow:

Start with training wheels:
→ ChatGPT Custom GPTs teach you how AI agents think
→ No setup required, works in your browser
→ Perfect for understanding custom instructions

Then graduate to real tools (for non-techies):
→ FlowGent for customer-facing agents (handles instagram dms, whatsapp, website chat)
→ Relay for internal workflows (massive template library, zero coding)

The secret? Don't build from scratch.

Every successful non-tech founder I know started with templates and pre-built solutions.

They learned the principles first. Then customized later.

Your first AI agent should solve one annoying problem. That's it.

What repetitive task would you love to never do again?

3 ai agent tools for non-techies

Building your first chat AI agent?

Here's the 6 basic (but critical) concepts you should know: (save this for later) 

Most people focus on making AI sound smart.
I focus on making AI DO things.

The difference? Revenue impact.

Here are the 6 core concepts you MUST understand ↴

1. Instructions (System Prompt)
→ The agent's brain and personality
→ Define WHO it is, HOW it behaves
→ Most people: "Be helpful"
→ Winners: "You're Alex, senior sales rep at CloudCorp..."

Take-away: be extensive, precise and granular in your prompting. For chat AI agents you need to add a lot of examples of old convos for the AI agent to be able to mimic your companies' tone of voice.

2. Knowledge Base
→ YOUR business information organized for AI
→ Not just uploading PDFs
→ Structure data so it's understandable for AI
→ Products, services, FAQs, policies

Take-away: organize your knowledge. Make sure everything is up to date. The AI agent won't know the difference between old and new data.

3. Memory/Context
→ Remembers entire conversation history
→ References what customer said 10 messages ago
→ Makes conversations feel human
→ Context is what creates connection

Take-away: in line with the knowledge base: the most important part of your AI agent is context. AI is nothing without that (or you will get generic AI slop).

4. Actions/Integrations
→ The agent's "hands" - what it can DO
→ Change subscriptions, update CRMs
→ Send emails, create tickets
→ Without actions = expensive chatbot

Take-away: an AI agent only becomes an agent when it can actually do things. In practice this usually means connecting your AI agent with an API.

5. Decision Logic
→ HOW agent decides WHEN to act
→ "Subscription change request" → update subscriptions
→ Makes decisions based on goals or based on direct instructions
→ ALWAYS give a fallback in case the AI agent can't reach it's goal

Take-away: there are usually two ways an AI agent makes it's decisions: based on goals ("You need to always solve the customer problem using the tools you have access to") or exact descriptions ("In case someone asks for a refund, use the "refund-tool"). Usually you will use a combination of both.

6. Platform/Channel
→ WHERE your agent lives and operates
→ Instagram DMs vs website chat vs WhatsApp
→ Each has different rules and expectations
→ Platform shapes behavior

Take-away: first choose the channel and goal, then start building the AI agent.

𝐓𝐡𝐞 𝐫𝐞𝐚𝐥𝐢𝐭𝐲: Most agents skip concepts 4-6.

They talk but don't act.

𝐍𝐨-𝐜𝐨𝐝𝐞 𝐭𝐨𝐨𝐥𝐬 𝐭𝐨 𝐬𝐭𝐚𝐫𝐭:
→ FlowGent (for chat AI agents)  
→ Relay/Lindy/Zapier (for workflwo AI agent)

Build agents that DO things, not just talk.


Repost ♻️ if you're ready to build agents that actually work.

It's time to start thinking like your AI agent.

Why?

Because most people build AI agents from their own perspective.

Then get confused when the agent makes weird decisions.

Here's what I mean:

Your agaent only knows what's in its 10K-20K token context window.

Everything else? It doesn't exist.

Try this exercise:

→ Take a task your agent needs to do
→ Limit yourself to only the context the agent sees
→ Try completing the task with that limited info
→ Notice where you get stuck

For example:

Imagine you're a computer use agent.

You get a static screenshot and poorly written instructions.

You click somewhere.

Then you close your eyes for 3-5 seconds while the action executes.

You open your eyes to a new screenshot.

Did your click work? Did you shut down the computer?

You have no idea.

This is how your agent feels every step.

What agents actually need:

→ Clear action descriptions (what each tool actually does)
→ Recommended next steps (guardrails for exploration)
→ Better context about what just happened

Debug like an agent:

When your agent makes a mistake, don't ask "Why did it do that?"

Ask "What context was missing that would have prevented this?"

Then feed the full agent trajectory back into the model and ask it directly:

"Why did you make this decision? What would help you decide better next time?"

The agent will tell you exactly what context it needed.

Stop building from your perspective.

Start building from theirs.


Repost this to help others understand the importance of context for AI as well ♻️

Claude for Chrome just launched (1,000 lucky testers only)

Anthropic just announced their browser-using AI extension.

That means that Claude can now see your browser, click buttons, fill forms, and navigate websites directly from a side panel.

But there's one important aspect to be very aware of with any browser AI:

Browser-using AI comes with serious security risks.

Anthropic ran 123 different attack scenarios and found a 23.6% success rate for prompt injection attacks.

Real example: A malicious email could trick Claude into deleting your entire inbox without asking permission.

Even with their current safety measures (site permissions, confirmation prompts, blocked risky sites), they only reduced attacks from 23.6% to 11.2%.

Still not zero.

My honest take: We need to be VERY careful with browser AI.

The potential is massive, but the risks are real.

That said, I'm genuinely excited to test this when I get access.

I'm curious how it'll compare to Perplexity Comet - the competition in browser AI is heating up fast.

Just don't expect access anytime soon. They're expanding "gradually" as they build confidence in their safety measures.

Have you already tried interesting use cases for browser AI?

Taplio

Carolina Posma's Linkedin Analytics

Who is engaging with Carolina

's Best Posts (last 30 days)

Want to drive more opportunities from LinkedIn?

Taplio

.css-1jgreyo{background-image:linear-gradient(284deg, rgba(39,43,72,1) 0%, rgba(33,47,183,1) 100%);color:transparent;-webkit-background-clip:text;background-clip:text;}Carolina Posma's Linkedin Analytics

Who is engaging with Carolina

's Best Posts .css-qqfgvy{font-size:var(--chakra-fontSizes-md);}(last 30 days)

Want to drive more opportunities from LinkedIn?

Carolina Posma's Linkedin Analytics

's Best Posts (last 30 days)