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).

Here's a quick tip on how to build an effective presence on LinkedIn.

Too many people on LinkedIn are "performing" for their current (and future) followers.

Instead of self-centered performing, focus on generous *sharing*.

We all have something to share that will be helpful/useful to others. Things we are experts in or passionate about. What's often even better is sharing mistakes we've made and lessons we've learned. 

Share, don't perform and you'll pull people in.

* I fall into this "performing vs. sharing" mode sometimes too.  

** "effective presence" is not just about the number of followers, likes or comments. It's about having genuine impact on somebody's thinking -- even if just for a moment.

Always assume good intent, but don't assume good intent always means good outcomes.

Well-intentioned people can be misinformed, misguided and make mistakes.

Example:  Me. 

I love HubSpot with my whole heart. My intentions are purely positive, but I make mistakes. Sometimes I'm wrong. 

The lesson I've learned: It's OK to give people the benefit of the doubt. If you don't feel strongly, let others have their way. But if you do feel strongly, don't resist pushing back on others just because their intentions are good. 

Factor in someone's positive intent (it matters), but don't over-index on it. Have the courage of your conviction.

undefined undefined

Lots of people much smarter than me have been talking about this compelling new idea called the "Context Graph". I love the idea, but am a wee bit skeptical of far away we are from it being something businesses can use.

A concept is making the rounds in AI circles right now that has many people very excited: context graphs. Foundation Capital published a piece calling

Open article: Context Graphs: Capturing The "Why" In The Age of AI by Dharmesh Shah • 4 min read

Context Graphs: Capturing The "Why" In The Age of AI

BREAKING NEWS: Anthropic launches Claude Cowork 

If you've been following me for more than 24 hours, you likely know that I spend many of my hours working in Claude Code. It's my primary tool for agentic coding. 

Claude Code is an agentic *system* that can be used for more than just writing code. In fact, they could have just called it Claude Agent. But, even though it can be used for doing things other than writing code, because it runs in a terminal (which is known to create anxiety amongst many), it's mostly developers or developer-adjacent people that use it.

But today, Anthropic wrapped up the kernel of Claude Code and exposed it through a new feature called "Claude Cowork". 



Claude Cowork runs inside of the Claude Desktop app (Mac), and does one super-simple (but powerful) thing: It gives you access to Claude Code capabilities within the friendly chat interface you already know.  Oh, and like Claude Code, you can let it have access to designated files/folders on your local system.

Given that I'm already a heavy user of Claude Code, Claude Cowork doesn't really change my life all that much -- but I think it's going to be useful for others. It'll let you prompt Claude to do things like organize your folders, create files (like a Word doc) or read-in a CSV of survey results -- and with some patience, let you do things in a browser.

Though useful, and a smart move by Anthropic to take their lead in the developer use case and extend it to more general personal productivity, its not as impactful as creating MCP a year ago. Claude Cowork is exposing existing functionality to local users of Claude, not global integration.

I'm more excited about the rumored updates coming to the the Sonnet LLM which could open up more capabilities for building agentic applications.   

NOTE: Claude Cowork is currently a research preview and only available to Claude Max subscribers.

The scariest three words to get as a response from an agentic coding app (like Claude Code):

"you're absolutely right". 

This often comes back quickly when I ask the agent whether a different approach wouldn't be better because it's approach wasn't consistent, it didn't follow the DRY (Don't Repeat Yourself) principle, etc.

What worries me is that if I'm so absolutely right -- then why make that mistake in the first place?  

This is why I try to build skills for my more common tasks (so it has more instructions and context for those things), but given the quality of the Opus 4.5, it surprises me sometimes the level of things it gets wrong -- even when working on the same task when the conversation is still in context (not compacted).

Anyways, we still live in magical times, but every now and then, feels like there's a glitch in the Matrix.

Current status:  Spending some quality time this weekend tinkering with OpenClaw (which was formerly Moltbot, which was formerly ClawdBot).

I generally don't obsess over optimizing my personal productivity. I'm doing this more to learn and add some "dots" to my collection so that some dots may connect in the future.

Some early thoughts:

* Ignore the fact that it's kind of hard to get working. That'll get better.

* Don't ignore the security risk. That'll get better too -- but even then, I'm worried. (Right now, I'm using it in a completely separate server with no access to my production files, email, etc.).  Even if we stipulate that the odds of this getting compromised are really, really low (which may or may not be true), the problem is that *if* it does get compromised and you're using it to anything near its fullest extent, that's a massive surface area of risk.    

* Be particularly careful with giving it access to your primary email account. Not because it can then read all your email, but because email is often used as the security mechanism for a bunch of different web applications. Imagine if a rogue agent got access to your email, it could just reset your password to most websites/apps (which usually just send a reset code/link to your email).

* Putting the security issues aside, the agent does perform remarkably well. And because it accomplishes things effectively via composition/chaining, there's a LOT it can do once you do give it access.  

* I really like that it's open source and running in an environment you control. That means you can take the memory/context it builds with you (should you choose to). You're not locked in.

* Conceptually it's not that dissimilar from Claude Code + skills/tools. But it's still primarily a text-based interface.  Will be interesting to see if it ends up some of the emerging UI protocols to embed apps/UI into a chat interface.  

* I'm interested in the emerging patterns around agents that run and do work in the background (i.e. don't have to be initiated/triggered by a prompt or command). 

* If you're working for a software company like I am (in my case, HubSpot) I don't think you need to invest (yet) in specifically supporting OpenClaw. Stay focused on providing robust API and MCP support.  It's possible that one day we'll want to support connecting OpenClaw (or things like it) into our products/platforms, but for now, it feels like MCP/skills is a simpler and cleaner way to accomplish that.

Now, back to more tinkering...and pondering.

An idea for the LinkedIn product team.

You know how we can currently at-mention companies (ex: HubSpot) -- which then links to the company's profile page?

It would be cool to have a lightweight way to configure *teams* within the company.  

Example, I would then be able to setup teams like execs (with Yamini Rangan, Jon Dick, Duncan Lennox, etc. within that team).  Or, I could set it up for a team for a specific product. group, project, geography, etc. 

But once setup, I could then at-mention a team like this:

Hey HubSpot/execs, tagging you here as you really should read this post by one of our customers on how they are using AI to drive more engagement -- and conversions. 

It's a simple idea, but would be really useful.  Sometimes you want to tag, thank or reference a group of people in a company, not the main company profile.

What do you folks think? Would this be useful?

I'm  a massive fan (and daily user) of agentic coding tools like Claude Code and Codex CLI.

One of the questions we'll need to ponder in 2026:

How do you make sure that the compounding value you are getting from the use of agentic coding exceeds the interest on the technical debt that AI creates along the way?

The hope is that we'll be able to use AI itself to pay off the technical debt as the models and tooling get better.

Just registered clawspot .ai (no website yet).

Here's the idea:

Setup a multi-tenant OpenClaw server making it easy for normies to setup an OpenClaw bot with core cloud-based features.

Thinking: GMail, Google Docs, Web Search, X/Twitter, long-term memory, server-based local working directory (to save files and such).

Would let people try out the power of OpenClaw with less of the headache of setup and such. 

Somebody's probably already working on something like this so going to sleep now (it's 2:30am) and see what kind of comments this post gets.

For convenience, will configure the domain clawspot .ai to redirect people back to this post for now.

We've long known that the best people have IQ + EQ.

When it comes to AI agents and their effectiveness, there's a new element — and it's multiplicative.

Agent Success = IQ  x EQ x  CQ

IQ = How well does it reason?
EQ = How well does it collaborate?
CQ (Context Quotient) = How well does it know YOU?

And because it's multiplicative, if an agent has a zero context quotient, it has zero value. (Apologies for revisiting grade school math).

Here's a quick thought exercise:

You can choose between two agents.

Agent A: 200 IQ. Bonafide Genius. Knows nothing about you.
Agent B: 150 IQ. Brilliant. Knows your business cold.

You pick Agent B. Every time. Why?

Agent A can solve PhD math problems.
Agent B can help solve YOUR business problems.

Raw intelligence without context is just confident guessing. (Also known as "management consulting"). 😀 

AI models know a lot about the world, but you need agents that know about YOUR world.

The real game isn't How smart is your model? It's How are you increasing the agent's CQ?

Imagine if it had...
...your top rep's call recordings.
...your angriest detractor's NPS comment.
...your happiest promoter's renewal notes.
...and everything else swirling around in your company

The agents that help you win aren't the ones with the highest IQ. They'll be the ones who know you best.

This is what we're furiously working on at HubSpot, because nowhere does context matter more than in GTM and nowhere is there more unified context than the HubSpot platform.  (Disclosure: As founder/CTO, I'm biased). 

We're excited about the progress the AI companies are making on delivering smarter LLMs  (higher IQ). We need that to build great agents. Meanwhile, we're working on delivering agents with higher CQ. (And stay tuned for the EQ part -- have thoughts on that too).

Anyhoo...

Key takeaway: In the age of AI agents...

IQ gets attention.
CQ gets results.

I wonder if Clawdbot has increased the number of KVM (Keyboard-Video-Monitor) switches being sold.

You know, so you can switch between your primary computer and the Mac Mini you bought, just so you can try out Clawdbot, and still sleep at night.

For those not yet familiar:  Clawdbot is an AI agent that you can give access to your computer (it runs locally), so it can pretty much do anything you can do on your computer.  Rumor has it a bunch of folks are buying separate Mac computers so they can try it out, but mitigate the risk.

Clawdbot is scary good. And by that I mean it's good, but scary.

Woo hoo! Shipping an early release of a new AI app I've been working on. (Look at me, shipping something in the middle of the day instead of 2am!)

If you're curious about premium website domains and their underlying prices and trends, I think you might enjoy playing with this app:

domainsales .ai (yes, I like simple domains).

You can  ask questions like:

 "what were the top .ai domains sold this year."  

"what was the average price for .org domains every year".

"what's the sales history for chat .com "

All feedback appreciated. Feedback is a gift.

Thanks.

-Dharmesh

NOTE: I'm using this (and other) AI apps to test out new Agent.ai platform features that allow you to use any vibe coding tool to build apps and integrate them into the marketplace. That way, you benefit from 2.5M+ users, shared context and a bunch of tools you can use in your agents.  More on al this next week.

The history of software is really the history of how humans interact with computers.

A (partial) history of user interfaces:

TUI: (Terminal User Interface): Type & Read

GUI (Graphical User Interface): Point & Click

MUI (Mobile User Interface): Touch & Swipe

VUI (Voice User Interface): Speak & Listen

CUI (Chat User Interface): Prompt & Respond

In the age of AI, here's where we're headed:

EUI (Emergent User Interface): Generate & Adapt

Where we're headed is having UIs be generated, fluid...dynamic.

UIs of the feature will be outcome-oriented. What's the best UI to show to *this* user to get to the outcome needed?

And, most importantly, like most things in AI, *context* is going to be super-important.

This is why I'm excited for the work we're doing around context at HubSpot -- and what creates opportunity for companies that deeply understand their customers/users.  

To create the best possible UI to deliver the best possible outcomes you need the best possible context.

An #idea for the agentic coding products like Claude Code, OpenAI Codex, etc.

Right now, the system (by default) shows me how long it worked on the prompt I just gave it. 

What would be even more useful is showing me the file count and number of lines added/changed/deleted.

Sometimes, I don't have the time to do a deeper review of the code changes, but I have in my head an approximation for what the change *should* have entailed. 

So, if I was thinking this particular request should involve about 3-4 files and ~100 lines of code, and the system comes back with 50 files and 500 lines of code, I know it took the wrong path.  I can then revert immediately, spend more time in "plan mode" and have it try again.

A relatively simple feature (I think), but would be useful -- even if it's turned off by default.

Was recently on a podcast with Manny Medina from Paid.

Was a great conversation around business strategy, pricing and of course, AI.

I'm not used to being on podcasts (not my natural habitat), but I think it turned out pretty well. Hope you like it.

Dharmesh Shah helped write the SaaS playbook.
Now he’s explaining why parts of it are about to get ripped out.

In today’s episode of Get Paid, the HubSpot co-founder doesn’t sugarcoat it:

AI agents aren’t a feature.
They’re a platform shift.
Seats still matter. For now.
Credits are inevitable.

And most “AI strategies” are just SaaS habits in disguise.
They go deep on why reasoning models changed the game, why vibe coding won’t save you, and why pretending this is a normal upgrade cycle is how incumbents quietly die.

If you’re building AI and telling yourself “we’ll figure pricing out later,” this episode is a warning shot.

🎙 New episode is live. Link in comments.

Woo hoo!  HubSpot's Service Hub moves to #1 on G2.

Congrats to the Service Hub team.

Sales Specialist | AI, Service Hub & Data Hub | Helping companies scale post-sales efficiency with data and artificial intelligence | HubSpot

Very exciting news!!! 🧡🔥
HubSpot Service Hub moves to the #1 position for Customer Success Software in G2!

More info here: https://lnkd.in/evQn3x4g

#CSP #SuccessPlatforms #ServiceHub #HubSpot

Nice to see folks combining the flexibility of Agent.ai and its access to a rich set of AI capabilities and data with the power of HubSpot Workflows for automation.

Senior Software Engineer | HubSpot-Focused Integrations & CRM Automation

I've been having a blast combining Agent.ai workflows with HubSpot automation lately, and the results have been impressive.

Here's what I've been able to automate:

Buyer Intent Detection -> Agent.ai monitors signals across multiple channels and automatically scores leads based on engagement patterns, content consumption, and behavioral triggers. When intent spikes, HubSpot workflows kick in with personalized outreach.

Company Intelligence Gathering -> No more manual research. The system automatically enriches company records with technographics, funding data, org structure, and relevant news. Everything flows directly into HubSpot properties, keeping our CRM current without lifting a finger.

Lead Qualification & Routing -> AI agents assess fit based on ICP criteria, then route qualified leads to the right team members with full context. Sales reps get notifications with a complete brief, not just a name and email.

Follow-up Orchestration -> Based on engagement signals, the system determines optimal timing and messaging for follow-ups. It's like having a sales ops team working 24/7.

The real power is in the combination: Agent.ai handles the intelligence gathering and decision making, while HubSpot executes the workflows and maintains the source of truth.

For teams drowning in manual research and qualification work, this stack has been a game changer.

Huge thanks to Dharmesh Shah for making this technology accessible and for his visionary presentation on AI at INBOUND this year. The future he outlined is already here, and it's incredibly exciting to build with these tools.

Want to automate your workflows with AI? Whether you're looking to streamline sales processes, enhance lead management, or build custom integrations, I'd love to help. Drop me a DM or comment below with what you're trying to solve.

#AIAutomation #HubSpot #SalesAutomation #AgentAI #CRMIntegration #MarketingAutomation #SalesOps #AIAgents #WorkflowAutomation #B2BSales #RevOps #SalesTech #AITools #BusinessAutomation #LeadGeneration #INBOUND2025

taplio

Dharmesh Shah's Linkedin Analytics

Dharmesh Shah

Performance ScoreDharmesh Shah's LinkedIn Performance

What is Dharmesh talking about?

Who is engaging with Dharmesh

Dharmesh Shah's Best Posts (last 30 days)

Ready to grow your LinkedIn presence?

taplio

.css-6ytdur{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;}Dharmesh Shah's Linkedin Analytics

Dharmesh Shah

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

What is Dharmesh talking about?

Who is engaging with Dharmesh

Dharmesh Shah's Best Posts .css-1emoo84{font-size:var(--chakra-fontSizes-md);font-weight:400;color:var(--chakra-colors-gray-600);}(last 30 days)

Ready to grow your LinkedIn presence?

Dharmesh Shah's Linkedin Analytics

Performance ScoreDharmesh Shah's LinkedIn Performance

Dharmesh Shah's Best Posts (last 30 days)