Dharmesh Shah

Founder and CTO at HubSpot. Helping millions grow better.

Career startup guy.

Author of the book &quot;Inbound Marketing&quot; (http://InboundBook.com)

Specialties: Entrepreneurship, Software Development, internet marketing,SEO

Founder and CTO

Founder of HubSpot, a venture-backed software startup.  HubSpot offers the industry's first inbound marketing system for small businesses.  The software is available as a hosted service (SaaS).

"Why should companies pay for SaaS (HR/CRM/ERP/etc.) when they could just vibe code them?"

I get variations of this question or comment with some regularity (granted, it's sometimes just me talking to myself). 

Here are some biased (but hopefully, well-considered) thoughts:

1) I am a big proponent and user of vibe coding (what I call "agentic coding"). I do it every day, 7 days a week, including Sundays. It's amazing.

2) My company, HubSpot is a software company. We have hundreds of professional engineers -- just about all of them use AI for product development too. They are brilliant and know how to build production-grade products.

3) Even with this powerful army of talent, the number of internal, core SaaS applications that we have replaced with a vibe-coded variant is exactly ZERO. The number of applications we plan to replace is also exactly ZERO. 

4) It's not the absence of talent that keeps us from rolling our own SaaS apps, it's the presence of focus. It would be silly to try and replace our HR, team collaboration, expense tracking and 100+ other SaaS apps we use when we can just buy them. Just doesn't make sense.

5) That's us -- as a software company at some scale. If you're a non-software company it makes even less sense for you. Doesn't matter how good the AI coding tools get. Let's say you *could* vibe code a replacement for that SaaS app you're using, who's going to maintain it? Who's going to keep up with industry trends? What are you going to do when the 20-something genius that vibe coded it over a weekend leaves the company?  Who do you call when there's a major bug? 

6) If you're a Fortune 500 company at some scale, perhaps you could pull this off for some discrete use cases and the tradeoffs are worth it. You have an IT/Engineering department that is larger than the population of some countries.  You can take on the pain in return for the positives.

For the millions of others, my advice is:

Spend every calorie possible on creating value for your customers.

Never spend more time debating the priority of a product improvement than it would take to actually go ahead and do it. 

Small improvements add-up and they're morale boosters for both customers and the team.

In the early times at HubSpot we had something called "daily delights". Basically, every day, the product team would *ship* at least ONE small improvement to the product that was noticeable by customers (and hopefully delight a few of them).

We're bringing a variation of that idea back whereby we find and fix those minor annoyances in the HubSpot product and drive some customer delight. I'm thrilled. 

Sometimes companies have a tendency to overthink things and undervalue the compounding value of seemingly inconsequential improvements.

BREAKING:  OpenAI launches GPT 5.2 and boy is it smart.

I got immediate value out of it.

I was struggling with getting Codex CLI to work through a tricky coding (actually deployment) issue I was running into.  I was going back and forth with the coding agent for almost a dozen rounds -- to no avail.

Was even desperate enough to try Claude Code with Opus 4.5.  No luck.

Then, saw the news about GPT 5.2.  Switched to it.  It took a couple of more rounds of prompting -- but it solved the issue.

Look forward to digging in more tonight with a late-night coding session for a major overhaul of one of my popular agents. 

Thank you OpenAI!

I build a lot of AI agents for myself, which I increasingly also share with others. 

I don't build AI agents to make people unnecessary.

I build them to make people unstoppable.

The latest McKinsey State of AI 2025 report just dropped and I'll cautiously say it's pretty good. (Cautiously, because I'm a bit leery of reports from big management consulting companies -- they're often very mega-enterprise focused and I'm a startup/SMB guy.)

Some quick takes and insights:

1. Most companies are still “AI-curious,” not “AI-capable.”

Eighty-eight percent are using AI somewhere, but two-thirds are still stuck in pilot mode. Translation: lots of glitter, not much gold.

2. AI agents are the new interns.

Sixty-two percent of organizations are experimenting with agents, but only 10 percent have them doing real work at scale. Early days—but the ones that figure it out first will have an army of tireless digital doers.

3. High performers think beyond cost-cutting.

The top six percent of companies don’t just use AI to save money—they use it to reinvent workflows and drive innovation. They’re not trimming fat; they’re growing new muscle.

My advice:
Experiment and play with AI. That's how you learn.
But don't stop there.
Lean in and truly leverage it.
Even if it's for small, discrete tasks.

The next frontier isn’t “using AI.”
 It’s working with AI.

At HubSpot we're investing in helping millions of SMBs make this transition -- just like we've done in the past. Everything from our featured agents like Customer Agent and Prospecting Agent all the way down to simple, task-oriented agents on Agent.ai.  

We're not interested in high-falutin', performative AI.  We focus on the practical: What can AI actually *do* for you?

What's the difference between a company's mission and vision?

Here's how I think about it...

A mission states who the company is, what it is pursuing and why.  It's describes what is (hopefully) happening *today*.

A vision describes where the company is going, what direction it's moving and where it aspires to be in the *future*.

For example, HubSpot's mission is to "help millions grow better". So, everything we do can be connected back to that mission. The products we build, the underlying platform it uses, our partner ecosystem, HubSpot Academy, our free tools, Agent.ai, our INBOUND conference...everything. It's all in service of that core mission of helping millions grow better. 

So, what's our VISION? Candidly, we're currently iterating on it. With the advent of AI, we think we can be more aspirational and set a higher bar in terms of what we can accomplish.

Trust Insights

I'm the resident AI person at Trust Insights. Got a marketing data, analytics, or AI strategy problem? I'll help you solve your mysteries.

I've built AI-powered attribution modeling (to know what's working and what isn't), predictive analytics for content marketing (to know what to publish and when), and digital customer journey mapping (to know what to say and where) - long before ChatGPT existed. In fact, I've been working in the AI field since 2013.

Today, I speak and teach around the world about how to make the most of the three classes of AI (including generative AI) for your business.

Co-Founder and Chief Data Scientist

Marketing Over Coffee

I'm the co-host of one of the top marketing podcasts, largely thanks to my talented co-host, John Wall. Each week we discuss old and new marketing, tips, tactics, methods, ideas, strategies, and much more to get more out of your marketing efforts. With over 700 episodes and a million downloads, it's one of the most popular marketing podcasts in the world.

Visit http://www.MarketingOverCoffee.com for more information and how to get the show for free.

Co-Host

Social Media Pulse

Member of the Social Media Advisory Board. I provide guidance to the team about social media trends, especially as they relate to marketing analytics, social media marketing analytics, social media and data science, and social media and artificial intelligence.

Board Member

The Order (Non-Profit)

The Order advocates for equitable treatment and representation of LGBTQ+ and female-led media and entertainment. Through strategic fundraising, public support campaigns, and advocacy, we aim to promote greater diversity, equity, and inclusion in the entertainment industry. Our vision is a world where all forms of media are celebrated and valued, regardless of creators' gender, sexual orientation, or background.

In my role as Head of Data Science for The Order, I:
- Led comprehensive data science initiatives, driving targeted advertising campaigns, analyzing advertising media landscapes, and conducting custom audience analysis.
- Developed and implemented advanced models and algorithms for audience inference, Twitter data processing, and propensity score risk assessment.
- Orchestrated the creation and management of custom software solutions for social media management, email marketing with integrated data, and media monitoring.
- Conducted extensive research on latent demand patterns, media consumption, and strategic acquisition opportunities, providing valuable insights for marketing strategies and decision-making.
- Produced captivating video ads and implemented innovative influencer marketing programs, expanding brand reach and driving engagement.
- Utilized predictive analytics to forecast demand patterns and develop propensity score risk models, optimizing resource allocation and decision-making processes.
- Crafted and executed public relations strategies, managing brand reputation and external communications.
- Oversaw survey data processing and reporting, generating actionable insights for marketing and business strategies.

Head of Data Science

SHIFT Communications

At SHIFT, I help clients understand how earned media and public relations impact their marketing efforts and demonstrating how to measure PR, marketing, and finding ROI in a big picture view. I run many of the inbound marketing and lead generation programs for SHIFT's own marketing, from CRM administration to Pardot marketing automation to Google Analytics and more. In the first 3 months, I exceeded my lead generation targets by 16x using proven, tested lead generation and content marketing strategies, implemented cutting edge data science and machine learning practices, and I've built a multimillion dollar business unit.

Learn more at http://www.shiftcomm.com

Vice President of Marketing Technology

National Speakers Association

I had the honor and privilege to address NSA about the future of public speaking, augmented by artificial intelligence. Feedback included:
"Your presentation was fantastic! Thank you for sharing a portion of your brilliance with us! #nsa18"
"@cspenn did a great job explaining how #AI will disrupt speaking! Wise speakers will take actions to leverage AI! #nsa18 #Speaker #disrupt"
"I loved the way you married great presentation style, data and analytics, and relevance to the group."

Speaker

Social Media Dallas

I had the honor and privilege to address Social Media Dallas about predictive analytics for social media marketers. Feedback included:
"Can you tell I was excited to be at #SMDallas? Thanks for the great presentation on predictive analytics @cspenn. "
"Thanks @cspenn and @SM_Dallas for a great presentation tonight!"
".@cspenn is blowing our minds teaching us about Predictive Analytics! #smdallas"

Keynote Speaker

Marin Software

Delivered the opening keynote at Marin Masters, New York City on cognitive marketing.

I spoke on the Rise of the Citizen Analyst, how the democratization of analytics tools is fueling the citizen analyst movement.

World of Watson 2016 Speaker

HubSpot

Delivered the track keynote on building a data-driven customer journey.

Track Keynote Speaker

Christopher S. Penn is an authority on analytics, digital marketing, marketing technology, data science, and machine learning. A recognized thought leader, best-selling author, and keynote speaker, he has shaped five key fields in the marketing industry: Google Analytics adoption, data-driven marketing and PR, modern email marketing, marketing data science, and artificial intelligence/machine learning in marketing. As co-founder and Chief Data Scientist of Trust Insights, he is responsible for the creation of products and services, creation and maintenance of all code and intellectual property, technology and marketing strategy, brand awareness, and research & development.

Mr. Penn is a 2024, seven-time IBM Champion in IBM Data and AI, a Brand24 Top 100 Digital Marketer, an Onalytica Top 100 AI in Marketing influencer, and co-host of the award-winning Marketing Over Coffee marketing podcast. Prior to co-founding Trust Insights, he built the marketing for a series of startups with a 100% successful exit rate in the financial services, SaaS software, and public relations industries. His work has served brands such as Twitter, T-Mobile, Citrix Systems, GoDaddy, AAA, McDonald’s, and many others.

Mr. Penn is an IBM Watson Machine Learning Certified Professional, a Google Analytics Certified Professional, a Google Ads Certified Professional, a Google Digital Sales Certified Professional, and a Hubspot Inbound Certified Professional. He is the author of over two dozen marketing books including bestsellers such as AI for Marketers: A Primer and Introduction, Marketing White Belt: Basics for the Digital Marketer, Marketing Red Belt: Connecting With Your Creative Mind, and Marketing Blue Belt: From Data Zero to Marketing Hero, and Leading Innovation.

Learn more about Christopher here:

https://www.ChristopherSPenn.com
https://www.MarketingOverCoffee.com
https://twitter.com/cspenn
https://www.trustinsights.ai

Important: If your first message after connecting is a pitch, then with regret I need to report your message as promotional/spam to LinkedIn and remove the connection.

Christopher Penn

Excited to announce that I will be attending and speaking at Social Media Marketing World once again this year.

Want to catch up in person? Let me know and we can chat.


#SMMW #SMMW2026

Co-Founder and Chief Data Scientist at TrustInsights.ai, AI Expert

NotebookLM got a HUGE upgrade. Several of them, in fact.

The two newest additions to the system include:
- an infographic maker
- a slide maker

As well as improved video show makers, and adding full Deep Research to the system.

And when you use the Deep Research tool, not only does it load the synthesis in as a source, it also loads all the sources it used into the notebook as sources, which means you get the original goods as well as the synthesis.

It's difficult to overstate just how useful these quality of life upgrades are to the tool. It was already a fantastic summarization and synthesis tool. Now it's a great generation tool as well.

But to get the most out of it, you have to remember one KEY principle:

It's Gemini under the hood.

I'll say that again: it's Google Gemini under the hood.

All the prompt frameworks and best practices you use with Google Gemini apply doubly so to NotebookLM. What separates good from great is the use of Gemini best practices. Prompting NotebookLM is the same as prompting Gemini.

For example, in the prompt for making infographics? Include your Gemini-optimized style guide, and you'll get infographics that are far more aligned to your brand style guidelines. 

You can use things like the Trust Insights CASINO prompt framework for research throughout the tool with incredible effect.

Here's one simple use case for you. Put ALL your emails and conference calls and work product from 2025 into NotebookLM. Then have it synthesize a performance review for you. Then convert that review into a source and have it build an infographic of your performance for 2025. You'll be amazed at how much you accomplished and perhaps your manager will be as well.

Go try the newly upgraded NotebookLM!

#AI #GenerativeAI #GenAI #ChatGPT #ArtificialIntelligence #LargeLanguageModels #MachineLearning #IntelligenceRevolution

Google is creating AI-generated search results now.

This is different than AI mode. This is different than AI overviews. This is a completely synthesized set of search listings, an experiment called Web Guide.

This evening I went to search for "phaser games audio library" to find something for a coding project, and I got these results.

Look at the notes in the footer:

"Headers and descriptions are AI generated and may include mistakes."

Now, apparently this was announced back in July, but this is the first time I saw it in my standard Google results. On my personal account, I've opted into pretty much anything experimental, but this is the first time I've seen it invoked - and I didn't specifically ask for it.

I have to wonder if the new Gemini launch included a big compute upgrade that lets them roll this out more widely.

What you'll notice is that the page titles and descriptions are different from the originals - Gemini is summarizing them to be more helpful from its perspective.

Think about what this means for content creation, when even the search listings themselves are intermediated by AI. All that SEO software you paid for to optimize snippets and page titles? Perhaps time to re-evaluate that if this experiment succeeds and Google sees better user satisfaction from what Gemini has to say about your site rather than what you have to say about it.

Over on Reddit, folks are commenting that Web Guide has been turned on in their accounts as well without asking for it specifically. Rand Fishkin Lily Ray Dale Bertrand Aleyda Solís Marie Haynes Joost de Valk Kevin Indig are you folks seeing an uptick in Web Guide rolling out?

Maybe the test is going bigger. Either way, this is a good hint about where Google thinks things should be going in its ecosystem.

#AI #GenerativeAI #GenAI #ChatGPT #ArtificialIntelligence #LargeLanguageModels #MachineLearning #IntelligenceRevolution

Plan big, act small has never been more important.

This is Principle 27 in my book, Almost Timeless: 48 Foundation Principles of Generative AI.

In the last month, we've had 3 MAJOR foundation model releases: GPT-5.1, Gemini 3, Claude Ops 4.5. These new models are incredibly capable. They're super smart.

They're also really compute and money expensive. If you've done more than 5 minutes of work with any of them and you're not paying top dollar for Max/Ultra/Bro plans, you've likely hit rate limits/model not available/come back later/pay me more notices.

The top mistake most of us make is using one model for everything, which burns up costly resources on trivial tasks. Instead of thinking of your favorite AI tool as a monolithic interface, a black box with one button, think of it like a consulting firm.

A senior consultant with a big billing rate shouldn't be doing intern or junior consultant work. 

A junior consultant should not be a project lead.

When you go to do a project, be it write a novel, build some code, design a website, etc., think like a manager. Delegate the hardest tasks to the senior consultant (expensive models like Gemini 3 Pro, Claude Opus 4.5, GPT-5.1 Thinking).

Once you've got your plan/outline/skeleton/framework, then delegate the implementation to a junior consultant (cheap models like Gemini 3 Flash, Claude Haiku 4.5, GPT-5.1 Mini). You will dramatically increase your productivity and not run up big bills, especially if you are paying as you go. (API vs. subscription)

On top of that, it boosts your AI sustainability footprint. You don't take a Harrier fighter jet to the grocery store, you take a car (or better yet a bicycle or even walking). Don't use the biggest models for everything - the smaller the model, the less of an environmental footprint you leave.

Again, think like a manager. Would you have your most expensive lawyer summarizing trial transcripts, or would you have your cheapest paralegal do that?

Plan big, act small.

Image generated by Gemini 3.

#AI #GenerativeAI #GenAI #ChatGPT #ArtificialIntelligence #LargeLanguageModels #MachineLearning #IntelligenceRevolution

Of all the models DeepSeek has released lately, none is more interesting to me than DeepSeek Math V2.

Not because I have a particular affinity for math. I really don't. It was the subject I did worst in school. (I credit my 5th grade teacher for instilling a deep seated lack of confidence in me by berating me publicly in class frequently for my present day math skills. Thanks, Mrs. Lalley!)

What makes DeepSeek Math V2 so interesting is that it's focused specifically on mathematical reasoning above all else. Where models like GPT-5.1-Codex focus on coding, DeepSeek Math focuses on proving theorems.

DeepSeek Math bests every model except Gemini 3 on ProofBench - in many cases handily. It's more than double the capability of OpenAI's GPT-5.

You might be thinking, I don't do a ton of theorem proofs day to day in my business job. Neither do I. We're not the audience for this particular AI model. What this AI model shows is a distinctly long point of view: building the tooling to build the tools.

An AI model that writes better content is good. An AI model that writes better code is great. But an AI model that writes better math unlocks entirely new fields of AI or optimizations we just haven't thought of yet. 

Recall way back in January when DeepSeek V3 came out and Silicon Valley lost a trillion dollars of valuation in a day. Why? Because in a single paper on GRPO, DeepSeek and similar Chinese companies showed that even with export restrictions, they could build better AI because they had better math.

Better AI comes from better math.

The trillion dollar data center build outs that have propped up the US economy in 2025? They are 1 math paper away from becoming obsolete before they're even built, because better math - better theorems especially - could cut compute costs by orders of magnitude. Suddenly the US government plan to subsidize AI looks like a giant money sinkhole rather than an avenue to the future.

And DeepSeek Math V2's architecture has some fascinating tricks up its sleeve - like the ability to automatically label new training data, effectively teaching itself how to do better math. It generates harder problems, then solves them, then verifies that it solved them, and then it trains on what worked and what didn't.

An AI that genuinely self-learns.

What this unlocks for a company like DeepSeek is the ability to generate smarter, faster, more powerful, and more efficient AI models than human mathematicians could build alone. This will increase the pace of change and, if other countries' AI labs don't keep up, enable Chinese AI companies to leap ahead ever faster in AI capabilities.

Better AI comes from better math.

#AI #GenerativeAI #GenAI #ChatGPT #ArtificialIntelligence #LargeLanguageModels #MachineLearning #IntelligenceRevolution

Imagine having to say the quiet part out loud to AI.

"Claude, your job is to screw the customer over as much as possible, maximizing profitability at all other costs, including customer happiness."

Associate Professor at The Wharton School. Author of Co-Intelligence

"Alignment for whom" is going to be a big question inside organizations as they deploy external-facing AI solutions...

(From Anthropic’s testing of Claude 4.5 Opus)

In this issue, I answer Chris Brogan's comments about AI use and explore how upgrading your vocabulary serves as the ultimate lever for better artificial intelligence results.
You'll discover why using specific industry jargon unlocks superior precision from large language models. You'll see how borrowing technical terms from software development forces AI to apply logical structure to your requests. You'll understand the critical difference between prompt recipes and data ingredients for generating accurate content. You'll master the art of context engineering to guide AI tools toward exact results rather than generic approximations.

#AI #GenerativeAI #GenAI #ChatGPT #ArtificialIntelligence #LargeLanguageModels #MachineLearning #IntelligenceRevolution

Almost Timely News: 🗞️ Better Vocabulary for Better AI Results (2025-11-30) :: View in Browser The Big Plug 🚨 Watch my latest keynote, How to Succes

Open article: Almost Timely News: 🗞️ Better Vocabulary for Better AI Results (2025-11-30) by Christopher Penn • 19 min read

Almost Timely News: 🗞️ Better Vocabulary for Better AI Results (2025-11-30)

Fastest way to spot AI content: when the bot powering it is completely unable to handle a context switch.

Last night on Xitter, I posted about Apple TV's new show, Plur1bus, and a bit of scientific speculation based on the episode. It is a complete context change from what I normally post about, which is marketing and AI. It talks about processing human bodies, among other things.

Sentences like this:

"A dairy processing plant could be retrofitted to have its centrifuges isolate blood serum instead of milk."

And someone replied immediately with:

"this intriguing concept raises important ethical conversations. How do we balance creativity with responsibility in storytelling? 📚 #ThoughtLeadership"

Bro.

Why does this happen? Some folks have their AI bots programmed to simply parrot and reply to anything posted by specific accounts. The language model - and it's undoubtedly something very fast, very light, and very cheap like Meta Llama 4 - jumps right in based on a rudimentary prompt, and you get... slop.

Hilariously irrelevant slop.

If you want to see who's human and who's not, try a context switch from time to time on the social network of your choice and see just whose bots completely miss the memo.

#AI #GenerativeAI #GenAI #ChatGPT #ArtificialIntelligence #LargeLanguageModels #MachineLearning #IntelligenceRevolution

taplio

The Battle

Dharmesh Shah

Christopher Penn

Followers

Posts

Engagements

Likes

Performance ScoreDharmesh Shah's LinkedIn Performance

Performance ScoreChristopher Penn's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Dharmesh Shah

Christopher Penn

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);}Dharmesh Shah's LinkedIn Performance

Performance ScoreChristopher Penn's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreDharmesh Shah's LinkedIn Performance