Tim Carden

Thoughtleader

The future of B2B belongs to companies that master systematic outbound.

We partner with high-ticket B2B companies to build predictable lead generation engines through cold email and LinkedIn outreach - data-driven campaigns that target the right people with personalized messaging that books meetings.

What makes us different:

1. Technology: Clay for enrichment, HeyReach for LinkedIn automation, multi-domain email infrastructure
2. Personalization: AI-powered message variations that avoid spam filters and pattern detection
3. Deliverability: Domain warming strategies and technical optimization for maximum inbox placement
4. Results: Every campaign tied to booked meetings and pipeline, not opens or clicks

We've engineered a system that identifies your ideal customers, reaches them with personalized messaging, and books qualified meetings predictably.

Our clients include:

• High-ticket B2B companies ($10K+ ACV)
• SaaS companies needing consistent pipeline
• Agencies and service providers selling premium offerings
• 7, 8, and 9-figure businesses ready to scale outbound

Interested?

Book a call with our team (link in featured section)

Co-Founder

Hey, I'm Tim, Co-Founder at Thoughtleader, and we solve the #1 problem killing B2B growth: empty sales pipelines.

We help high-ticket B2B companies flood their calendars with qualified meetings using systematic cold email and LinkedIn outreach - data-driven campaigns that target decision-makers and book meetings on autopilot.

We can help you if:

• You're a B2B founder/CEO who needs predictable pipeline NOW... We'll deliver qualified meetings within 2-4 weeks using proven outbound systems

• You're scaling a 7-9 figure business and tired of inconsistent lead flow... Our cold email + LinkedIn campaigns run 24/7, filling your calendar while you sleep

• You sell high-ticket solutions ($10K+ ACV) and need buyers with real authority and active need... We target the right people with personalized messaging that gets responses

Our approach:

• Clay for data enrichment and hyper-personalization
• HeyReach for LinkedIn automation at scale
• Multi-domain cold email infrastructure for maximum deliverability
• AI-powered message variations to avoid spam filters
• Systematic qualification to ensure meeting quality

We've cracked the code on B2B outbound that actually drives revenue. Predictable meetings, qualified prospects, scalable systems.

If you:

• Sell high-ticket B2B solutions (>$10k contracts)
• Need qualified meetings, not vanity metrics
• Are ready for predictable, scalable lead generation

Let's talk.

Book a call 👉 https://thoughtleader.typeform.com/workwithus?utm_source=linkedin

Or DM me directly and I'll show you exactly how we're getting these results.

We analyzed our cold email costs and found a $2,600/month inefficiency. Here is the exact infrastructure switch we made to fix it.

We stopped buying pre-warmed inboxes from Instantly.

Great tool, but the math for pre-warmed domains doesn't scale for us.

Here is the breakdown.

Instantly requires 5 inboxes per domain. To keep deliverability safe (under 60 emails/day per domain), you are capped at 12 emails per inbox.

You pay $50/domain to send 60 emails.

We switched to ZapMail.

They allow 3 inboxes per domain. We send 20 emails per inbox.

Same 60 emails/day output. But it costs $24 instead of $50.

At 100K emails/month, the difference is massive:

• Instantly pre-warmed: $3,000
• ZapMail generic: $1,400
• ZapMail custom: $400

That is $2,600 in monthly savings just by optimizing structure.

Here is our new playbook:

Day 1: Client signs. We launch immediately on ZapMail generic pre-warmed domains.

Week 1-4: We buy custom domains and warm them in the background.

Week 5: Pivot to custom on-brand domains.

We get the speed of pre-warmed inboxes without the inflated cost.

And for anyone worried about "generic" sender domains:

We book 20+ meetings a week using them. Prospects care about the offer, not the domain name.

Don't let infrastructure costs eat your margin. Run the math.

Co-Founder @ RevenueFlow. Filling calendars with qualified meetings for high-ticket B2B companies through cold email

No alternative text description for this image

I built a custom lead database of 4,000 companies for pennies.
Most people are stuck paying $99/mo for limited export credits. Or they are manually copying data from Google Maps.

I just built a custom scraping engine in 15 minutes without writing a single line of code.

The "High Agency" Stack:

Cursor (The Builder): The AI code editor. It writes and runs the Python script.

1. Serper.dev (The Source): Scrapes Google Maps results (e.g., "Private Schools in Wyoming").
2. Firecrawl (The Scraper): Visits every website and turns it into clean Markdown.
3. OpenRouter (The Brain): Uses cheap LLMs to extract decision-makers from that Markdown.

The Workflow:

I opened Cursor and typed:

"I want to find private schools using Serper.dev. Then use Firecrawl to visit their sites, and OpenRouter to find the Principal's name."

Cursor planned the steps. Cursor wrote the code. Cursor fixed its own errors.

The Result: A CSV with 4,000 rows of enriched data, ready for Clay.

The barrier to entry isn't "Coding" anymore. It's curiosity.

If you can combine APIs, you can build anything.

How I pulled a list of 4,300 verified U.S. banks in 2 minutes without writing a single line of code.

Normally, getting needed data from an API means one of two things: Waiting days for an engineer to help or giving up and doing it manually.

Not anymore.

Here is the exact workflow I just used to get this done over a cup of coffee:

1. I signed up for the bank data API provider and got my key.
2. I downloaded their documentation.
3. I opened Cursor (the AI code editor).
4. I pasted the API key, the docs, and my plain English request into the chat.

That’s it.

Cursor read the docs, wrote the script, and ran it for me.

Two minutes later, I had a clean CSV file with 4,300 active U.S. banks, complete with website URLs.

No coding. No debugging. Just the data I needed, ready for enrichment in Clay.

The barrier between "technical" and "non-technical" is gone.

If you can clearly describe what you want, you can get it.

Stop saying "I can help your company find more customers"
It’s lazy. It’s generic. And it gets deleted.

Start saying: I can pitch your 10-week leadership accelerator to 2,000 CEOs and founders.

The difference in response rate is astronomical.

Why? Because you are taking their specific product and putting it directly into the copy. It proves you know who they are without doing gimmicky I saw you went to this university personalization.

Here is the exact 2-step workflow we are building right now to automate this at scale:

Step 1: The flagship column in Clay 

We are building an AI web scraper in Clay that visits a prospect's website with one goal:

• Find their flagship program, service, or product name.
• Return that exact name into a column in our table.

Now, instead of a generic pitch, our variable inserts their actual core offering into the sentence.

Step 2: The testimonial PS We are adding a second AI column that scrapes their site for customer testimonials.

If they have a glowing review from Chris the CEO about saving 50% on taxes, our email now ends with:

PS: saw Chris's testimonial about how you cut his tax costs by 50%. That's exactly the kind of leader we reach.

This isn't magic. It's just better data applied intelligently to copy.

Don't send generic emails. Use AI to find the specific details that make your offer a no-brainer.

If you want to see the exact prompt I used to get Claude to suggest these changes, let me know in the comments.

Manual filtering is a waste of human capital.

Clicking "Location," then "Industry," then "Headcount" for the 100th time is not high-leverage work.

The goal is to get the data into the enrichment table as fast as possible.

Here is the "Sourcing Waterfall" I use to keep my team efficient:


Level 1: Clay People Finder (The Default)

We already pay for the subscription.

The "Sculptor" feature allows you to type natural language queries like:

"B2B SaaS companies in Boston with 11-50 employees."

The AI automatically maps this to the correct filters.

You verify the output, press import, and you are done.


Level 2: Apollo via Cursor

If Clay doesn't have it, we go to the big database.

We connect via API (using Apify or RapidAPI) inside a Cursor workflow.

This handles volume better but requires slightly more setup than Level 1.


Level 3: Serper (Local Business)

If the target is offline (plumbers, gyms, restaurants), databases often fail.

We use Google Maps scrapers via Serper to find these.


Level 4: Recursive Search (The "Hard" List)

This is for "Lookalikes."

We query Google for "Top 10 competitors to [Company X]."

Then we take those 10 results and find their competitors.

This builds a massive, highly specific list from scratch.

The Lesson:

Don't over-engineer the scrape.

Start with the tool that speaks your language.

Only escalate to code when the easy button breaks.

I just scraped 2,000 leads from Apollo in 5 minutes. 
Total cost: Roughly $0. Lines of code written by me: 0.

I didn't use a complex enterprise tool. I used Cursor (AI) + RapidAPI.

Here is the workflow:

1. Find the data: Get a search URL from Apollo.
2. Find the pipe: Get an API key from RapidAPI (search "Apollo Scraper").
3. Command the AI: Tell Cursor, "Connect this API to this Search URL and give me a CSV."

If you hit an error? You don't call a developer. You screenshot the error, paste it back into Cursor, and say: "Fix this."

Why this matters: We are moving from an era of "Learning to Code" to an era of "High Agency Problem Solving."

If you can articulate the outcome (I want leads) and the inputs (Search URL + API Key), the AI builds the bridge.

Don't let "I'm not technical" be your excuse anymore.

I automated my entire cold email infrastructure in 15 minutes.
Buying inboxes, verifying settings, and managing warmups.
Here is the Self-Healing inbox workflow:

We just solved the biggest pain in cold email: The Ramp Up.

Buying pre-warmed inboxes is easy.

Safely transitioning them to full volume is hard.

The "Standard" Way:

• Buy 50 inboxes on Zapmail.
• Manually check creation dates.
• Set limit to 5. Wait a day.
• Set limit to 10. Wait a day.
• Set limit to 17.

That is 150 manual adjustments over 3 days.

We automated it with one script.

The Stack: Zapmail + Instantly + Cursor.

1. The Setup

Buy the pre-warmed inboxes on Zapmail.

Use Cursor to write a script that automatically exports them to Instantly and tags them "Zapmail Pre-Warmed."

No CSV exports. No manual entry.

2. The Ramp Up Logic

We wrote a simple script that runs every 24 hours.

It checks the "Creation Date" of every inbox with that specific tag.

• If Age = 0 Days -> Set Limit to 5.
• If Age = 1 Day -> Set Limit to 10.
• If Age = 2 Days -> Set Limit to 17.

3. The Self-Updating Documentation

I added a "Rules File" to the project.

Every time the AI learns a new API endpoint or a better workflow, it updates its own instructions.

The system gets smarter the more we use it.

You don't need to be a developer.

You just need to be tired of doing it manually.

How we turned a dead prospect list into 22,000 targeted leads in under 10 minutes. We had a list of 2,200 private schools. The goal was simple: find the decision-makers.

But our initial searches were dry.

The problem? We were using standard corporate language for a non-corporate niche. You won't find many "Chief Revenue Officers" at a K-12 private school.

We had to throw out the B2B playbook and learn their actual language.

Here is the iteration process we used in Clay to find the people:

1. The obvious titles. We searched for Principal, Head of School, and Assistant Principal. Result: ~1,800 prospects. (Okay, but poor coverage for 2,200 schools).

2. The department heads. We added Director of Admissions, Marketing, and Technology. Result: ~2,100 prospects. (Better, but still shallow).

3. The real decision makers. We stopped guessing and looked at actual school org charts. We added two huge titles: Executive Director and Superintendent.

Result: The list exploded to 22,000 prospects.

If your data enrichment isn't working, it’s rarely the tool's fault. It's usually because you don't know the target's specific job titles.

Don't force your language onto a niche industry. Learn theirs.

Why we buy 3x more email infrastructure than our clients actually need.

It seems wasteful. Why pay for 30,000 sending capacity if the client only has 10,000 leads?

Because deliverability is a game of margins.

If you only have "just enough" domains, you are forced to keep using them even when their reputation starts to crack. You grip the wheel tighter as you slide into the spam folder.

When you have 3x capacity, you have the luxury of rotation.

• Health score hits 98%? Rotate it out.
• Open rates dip for 2 days? Rotate it out.
• Need to double volume tomorrow? You have the runway.

We also cap every single domain at 40 sends per day. No exceptions.

Conservatism is the new aggression in cold email.

If you want to land in the primary inbox in 2024, you need to over-engineer your infrastructure.

The perfect lead list isn't bought. It's built.
But most sales ops teams struggle with the trade-off between Volume and Accuracy.

If you want high volume, the data is usually bad.

If you want high accuracy, the data is usually expensive.

The solution is not one tool. It is a "Waterfall" strategy.

We just ran a test on 10,000 leads.

Total cost: $40.

Here is the stack and the logic:

1. The "Free" Layer (Clay)

Always start here.

We use the Clay People Finder because it is already included in the subscription.

Max out your free/included credits first before paying for external calls.

2. The "Filter" Layer (Python + RapidAPI)

When you need raw volume (30k+ rows), use RapidAPI (Apollo).

The data comes back messy.

Instead of paying a tool to clean it, use a simple Python script to filter the list locally.

Why: You can run parallel analysis to process thousands of rows instantly without hitting API rate limits.

3. The "Deep Scrape" Layer (Apify)

This is the heavy hitter.

Apify acts like a real user. It mimics a browser, clicks through pages, and scrapes data that APIs often miss.

The Cost: ~$1 for 250 leads.

The Result: 10,000 highly accurate leads for $40.

The Workflow:

1. Check Clay for easy matches.
2. Pull the rest via RapidAPI.
3. Use Python to filter the noise.
4. Send the high-value targets to Apify for deep enrichment.

You don't need a $20k data budget.

You need a better workflow.

We realized we were burning money on simple classification tasks.

Using GPT-4o or even 4o-Mini to check if a lead is a "General Manager" is overkill.

It is like hiring a PhD to make coffee.

You don't need maximum intelligence for basic filtering. You just need basic comprehension.

So we moved our workflow from Clay's native OpenAI integration to OpenRouter.

The math is simple.

Native GPT-4o-Mini is cheap.

But models like Mistral or open-source 120b models via OpenRouter are 20% of that price.

When you are running 10,000 rows, that difference pays for your entire software stack.

Here is the workflow to set it up:

The Stack: Clay + OpenRouter + HTTP API

1. Get an OpenRouter API key
This gives you access to every model (Llama, Mistral, Qwen, etc.) with one credit balance.

2. Create a "Saved Enrichment" in Clay
Don't use the "AI" column. Use the "HTTP API" integration.

3. Configure the Request
Method: POST
URL: https://lnkd.in/eVSE2FNW
Headers: Authorization: Bearer [Your_Key]

4. Select the "Economy" Model
In the JSON body, set your model to something lightweight like `mistralai/mistral-nemo` or `liquid/lfm-40b`.

5. Prompt for JSON
Ask the model to return a structured JSON object so you can parse the answer easily in Clay.

The result?««

We get the exact same filtering accuracy.

We have access to models that aren't even listed in Clay yet.

And our API bill dropped by 5x.

You don't need to be technical to do this. You just need to stop paying the "convenience tax" on default settings.

Stop using Anthropic inside Clay. It is burning your credit balance for no reason. Here is how to switch to OpenRouter and cut costs by 80% (without losing quality):

The Problem:

Most people default to the "Smartest" model for every single column.

They use Claude or GPT-5 Mini for simple tasks like:

• Normalizing company names
• Formatting dates
• Cleaning up job titles
• Categorizing leads

This is like hiring a PhD to make coffee.

The Solution:

Use Open Source models via OpenRouter for any task that does not require the internet.

If you just need to format text or use internal logic, Open Source models (like Llama or Mistral) are just as good.

And they are significantly cheaper.

The Math:

• Anthropic/OpenAI Mini = Expensive for bulk ops
• Open Source via OpenRouter = 80% cheaper

The Workflow:

1. Open your Clay table
2. Click "Add Enrichment"
3. Select "OpenRouter" (instead of standard AI)
4. Choose a high-performance OSS model (like Llama 3 or Mistral)
5. Paste your prompt

The Rule of Thumb:

Does this column need to browse the web?

• Yes → Use GPT-5 Mini
• No → Use OpenRouter OSS

You get the exact same result.

But your credits last 5x longer.

I ran an accidental experiment that completely broke my beliefs about cold email infrastructure.

For years, the standard playbook was simple.

Buy a domain that looks like your main brand (e.g., getrevenueflow.com).

Warm it up for 3 weeks.

Launch campaigns.

But recently, our deliverability started tanking.

Replies were low. Open rates were inconsistent.

So we decided to test something radical.

We scrapped the on-brand domains entirely.

Instead, we bought "Generic Pre-Warmed" domains.

These are random domains, sent from random names, that have been warming up for 9-10 weeks.

The logic?

Deliverability is the only metric that matters.

If the email lands in the Primary inbox, you win. If it lands in Spam, the perfect offer doesn't exist.

We ran the test on 1,400 emails.

The results were outrageous.

4.2% reply rate (compared to our usual 1%).

19 sales opportunities generated.

A 5x spike in website traffic immediately after launch.

Why did this work?

1. The Warm-Up Period

Our standard domains get 3 weeks of warm-up. These generic ones get 10 weeks.

Google and Microsoft trust history. A 10-week-old domain with consistent traffic looks legitimate. A 3-week-old domain blasting volume looks suspicious.

2. Pattern Matching

When you buy 50 domains that all use the Founder’s name, AI spam filters catch on.

"Why is Tim sending 50,000 emails from 50 different URLs today?"

Generic domains use random names. There is no pattern to flag.

3. The "Trust" Fallacy

We assume prospects check the sender address meticulously.

They don’t.

They check the offer. If the offer is good, they reply. If they are curious, they click the link in your signature (which redirects to your real site).

Nobody cares if the email came from "John at GenericService.com" if the problem you solve is urgent.

The Operational Upside

Beyond the reply rate, this fixed our cash flow.

We no longer pay for "dormant capacity."

We don't have to buy domains and sit on them for a month while paying the monthly fee.

We treat domains like inventory.

Need to scale up? Buy pre-warmed domains instantly.

One burns out? Replace it instantly.

The Stack: Zapmail (for pre-warmed inventory) + Instantly (for sending).

Stop optimizing for "looking professional" in the headers.

Optimize for landing in the inbox.

Speed to lead is the highest leverage metric in sales. But drafting replies manually is a bottleneck. And fully autonomous AI often hallucinates or sounds robotic. The solution? 

"Human in the Loop" AI Agents.

Here is the exact workflow to automate 90% of your inbox management using Instantly.

The Stack: Instantly (Reply Agent) + Slack (Notifications).

1. The Setup

Go to AI Agents in Instantly and select "Human in the Loop".

Never use "Auto-Pilot" for sales. You cannot afford a hallucination sending the wrong pricing or a weekend meeting time.

You want the AI to draft the email, but you want final approval.

2. The "Brain" (Guidance)

Treat the "Guidance" section exactly like you are training a new sales rep.

Don't just say "be professional."

Be specific:
• "Suggest 3 times between 1-5 PM EST (Weekdays only)."
• "If they ask for pricing, refer to the attached PDF."
• "Keep replies under 50 words."

3. The Optimization Loop

When the AI drafts a reply, you get a Slack notification.

You open the inbox. The draft is waiting.

If it’s perfect? Click Send.
If it’s wrong? Edit it manually AND update the guidance.

Example: The AI suggested a Saturday meeting.
Fix: Add "Never suggest weekends" to the guidance doc.

In 2 weeks, your agent will be smarter than most junior SDRs.

The Result:
No more copy-pasting context.
No more "checking the calendar."
Just "Review, Approve, Send."

You move from "doing the work" to "managing the agent."

Most people structure their Clay prompts like this:
"Write a personalized intro for {{Company Name}} based on their recent news about {{News Topic}} using the data from {{LinkedIn Bio}}."

This works fine until you need to fix it.

If the output is bad, you usually want to take that prompt to a smarter model like Claude 3.5 Sonnet or Gemini 3 to iterate.

But you can't.

Because specific variables like {{Company Name}} are woven into the sentences.

You have to manually un-weave and re-weave them every time you paste back and forth between Clay and your external LLM.

It kills your iteration speed.

Here is the fix.

Group all your inputs at the very bottom of the prompt.

Structure it like this:

1. Context (Who you are)
2. Task (What to do)
3. Constraints (Length, tone)
4. Inputs (The variables)

It looks like this:

---

[The Logic]
Analyze the provided news and write an intro.

[The Inputs]
Company: {{Company Name}}
News: {{News Topic}}
Bio: {{LinkedIn Bio}}

---

Why this works better:

• You can copy the top 90% of the prompt into any LLM to refine the logic.
• The variable mapping at the bottom never breaks.
• Modern models are smart enough to read the context at the end.

The Stack: Clay (Execution) + Gemini 3 (Iteration).

Don't let messy syntax slow down your build speed.

Isolate the logic from the data.

Alejandro Dinsmore

Zippy

Staff Product Manager

Senior Product Manager

Founding Product Manager - 1500% increase in Loan Volume in 2023. 

Working on B2C and B2B initiatives across the business.

Product Manager

Thrive

Founding Product Manager for Thrive Pulse - 0 to 30,000 WAUs in six months

Product Manager for:
- Thrive Pulse
- Thrive for Chat (Slack and Teams)

THE ENDLESS CO.

Managed projects ranging from a mobile app for Covid-19 diagnostics to an augmented reality sneaker activation. Helped prepare pitch materials for new projects and clients.

Strategy and Product

eevo, Inc.

Built and launched virtual reality experiences for customers like Merck, Zoom, Logitech, McKinsey and The BBC.

AlphaSights

I worked predominantly with management consultancies and institutional investors - communicating daily with senior professionals and executives in early stage industry research, due diligence, and M&A processes, worldwide across all industry sectors.

Analyst

Sisasa Inc.

MIT D-Lab

Funded by MIT D-lab, worked specifically on solarclave project in Nicaragua.

Solarclave Medical Device Intern

Finding ways to live a better life through intentional work.

Founder (raised 2M in funding) | Early-stage Product Manager  (0-9M ARR) | Forbes 30u30 (not in jail)

Gemini 3 was just released! There was a lot of hype and rumors but I was not expecting Google to 2x the previous best score on ARC-AGI 2 (historically one of the toughest benchmarks out there)

Here's a bit more context on why this is a big deal:

ARC-AGI-2 is designed to stress test the efficiency and capability of AI reasoning systems, and provide useful signal towards AGI.

Pure LLMs score 0%

AI reasoning systems score only single-digit percentages (until recently)

Google's Gemini 3 just scored 45%, more than doubling the previous best of ~18% (from OpenAI's Pro model)

For all the "LLMs are plateauing" "New models aren't getting better", feels like progress is only accelerating.

Staff Product Manager | Fintech & Ai | Forbes 30 under 30

Wow Google is on FIRE this week. They just released a major update to NotebookLM and the new slide functionality is incredible!

Turn your sources into a detailed deck for reading OR a set of presentation-ready slides. These are fully customizable too so you can easily control the look, feel and content.

Check out this presentation it made from one of my sources. 

It took two minutes and for educational purposes I'm posting the version it created with zero edits.

What do you think? Does it look like slop or is AI finally getting good enough to handle building out decks?

Creatingyourfirstagent

Harnesses are the new application layer for AI.

Just recently Poetiq became the first team to surpass the 50% accuracy barrier on ARC‑AGI‑2 and they did it by building a Harness for Gemini 3. 

They used the exact same LLM, all they did was update the scaffolding and steps the model takes to solve the problem set. 

This refinement (which they open sourced) improved performance on ARC-AGI-2 from baseline 31%, $0.81/task up to 54%, $31/task.

That's a HUGE jump in performance and efficiency just from providing better scaffolding. 

I've been really excited about the opportunity to build scaffolding around existing models so it was cool to see a team execute on this opportunity so effectively. 

A harness I've been experimenting with for a while is one to help make AI writing sound more natural, check out the video below for a full overview and intro to harnesses and how you can make your own using tools like OpenRouter, Inc.

Using NotebookLM's vanilla (no prompt) infographic function vs adding a custom prompt makes a HUGE difference in quality!

No prompt (L)
Custom prompt (R)

(I'll drop the prompt in the comments)

This is just one example, and you can customize the infographic creation however you want (with the rights prompts) but if you're not using NotebookLM you're missing out.

Starting to wonder what interviews will look like in 12 months.

Poetiq scored higher than Google on ARC-AGI II and they did it using Google's own model

Google solved these visual puzzles with Gemini 3's unmodified test-time reasoning (costing $77 per problem)

Poetiq beat them (54% vs 45% accuracy) by building a better harness around Gemini 3 that scored higher but was also cheaper at $30/task

They added intentional scaffolding for the LLM to write a Python script to solve the grid instead of predicting the answer textually. 

Since the output is code, it's verifiable. 

The system runs the generated script against training examples, catches bugs, and lets the model fix its own code locally before committing to a final answer.

The exciting part is that harnesses like these are generalizable. Methods that work to improve Gemini 3 performance on ARC AGI can be applied in other domains. 

This is the new frontier.

Poetiq's Approach

It used to be easy to just pick the top model on the leaderboard and move on. That approach is effectively dead.

We’re looking at three distinct specialists now. 

You’ve got Gemini 3 Pro winning on general chat vibes and vision, Claude Opus 4.5 handling the heavy lifting for shipping actual code, and GPT-5.2 taking the lead on abstract reasoning and hard math.

Each model's strengths reflects the overall hypothesis and strategy that Google, Anthropic and OpenAI are taking in this space.

Google is investing in multi-modal and visual understanding. 

Anthropic is going all in coding and enterprise.

OpenAI is races towards a model which can produce novel research insights on its own.

I put these slides together (with NotebookLM's help!) to map out exactly where each one fits. It’s less about which model is "better" generally, and more about which one you actually need for the task in front of you

Which model is best for you?

OpenAI just dropped a 20 page guide on "Building an AI-native
engineering team". 

It's worth reading the full thing, but I've also been really loving using the new Google image model to create visual summaries of articles to keep for future reference.

Here's an example output based on the full 20 page guide:

taplio

The Battle

Tim Carden

Alejandro Dinsmore

Followers

Posts

Engagements

Likes

Performance ScoreTim Carden's LinkedIn Performance

Performance ScoreAlejandro Dinsmore's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Tim Carden

Alejandro Dinsmore

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Tim Carden's LinkedIn Performance

Performance ScoreAlejandro Dinsmore's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreTim Carden's LinkedIn Performance