How do you compete with AI and win? In this TEDxBoston talk, HubSpot co-founder and CTO Dharmesh Shah argues that the question is usually misread. Competing against AI is a losing game; competing with it, as a tool you use better than others do, is how individuals and teams come out ahead. His summary line: AI will not take your job, but someone using AI might, and AI may well take your job and hand you a better one.
Key takeaways
- Read "compete with AI" as "compete using AI". The people at risk are those who refuse to use it, not those whose work it touches.
- Banning AI at work or in school is, in his framing, like refusing spellcheck: the tool is here, and the skill is using it well.
- Natural language is becoming a programming interface. Shah describes his teenage son "programming in English".
- The shift is from "you versus AI" to "you to the power of AI": the value is in what a person plus AI can do together.
- For engineers, this is already happening in chip design, where AI-assisted EDA flows are part of production tool chains.
About the talk
| Item | Detail |
|---|---|
| Speaker | Dharmesh Shah, co-founder and CTO of HubSpot |
| Event | TEDxBoston (an independently organised TEDx event) |
| Published on YouTube | 2 July 2025 |
| Length | About 15 minutes |
| Theme | How individuals and businesses can work with AI rather than against it |
Shah co-founded HubSpot, the customer relationship and marketing software company, in 2006. According to the talk's description he is also an angel investor in more than 120 companies and writes the simple.ai newsletter, which the description puts at over 1.7 million subscribers. He was an early AI adopter: he says he first used GPT about two years before ChatGPT launched, when it had no user interface.
The main ideas, in brief
The points below follow the structure the talk itself sets out. They are summarised in my own words; watch the video for his examples and delivery.
1. The question most people get wrong
When people hear "how do you compete with AI", most picture a contest: human against machine. Shah's point is that the more useful reading is competing using AI. How you read the question shapes what you do next: defend your current tasks, or learn to do bigger ones with help.
2. How AI works, without the jargon
The talk sets out to explain how AI works without technical jargon. In plain terms, a large language model has learned patterns from very large amounts of text and predicts what comes next. The aim for most people is not technical depth but enough understanding to rely on it where it is strong and check it where it is weak.
3. Banning AI is like refusing spellcheck
Organisations and schools that ban AI tools, he argues, are making the same mistake as refusing spellcheck or calculators. The tool will be used anyway; the real question is whether people learn to use it well and responsibly.
4. "Generation AI"
Watching his 14-year-old son build software by describing what he wants in English, Shah sees a generation that treats AI as a normal working partner. Their advantage is not technical knowledge but the absence of a habit of doing everything by hand.
5. From threat to superpower
The talk closes on a simple shift in mindset: stop measuring yourself against AI, and start measuring what you can do with it. "You to the power of AI" is his phrase for that multiplier.
What this means for semiconductor and hardware engineers
Shah speaks mainly to business and software audiences, but the argument carries over to chip design, where AI is already part of the tool chain. Synopsys and Cadence both ship AI-driven optimisation for physical design and verification (Synopsys.ai and Cadence Cerebrus and JedAI, for example), and LLM assistants are being used to draft scripts, testbenches and documentation.
| Engineering task | Competing against AI | Competing with AI |
|---|---|---|
| Writing a testbench or scripts | Hand-writing all boilerplate | Generating a first draft, then reviewing and correcting it |
| Design space exploration | Manual sweeps of a few settings | AI-guided sweeps of many settings, engineer sets goals and checks results |
| Debugging a regression | Reading every log line | Using AI to cluster failures, engineer confirms the root cause |
| Learning a new protocol | Reading the whole specification first | Asking targeted questions, then verifying against the specification |
The common thread is that the engineer keeps ownership of correctness. AI drafts and searches; people decide and sign off. In hardware, where a silicon bug can cost a respin, that review step is not optional.
Practical steps
- Pick one repetitive task this week, such as a script, a report or a first draft, and do it with an AI assistant. Measure the time saved and the corrections you needed.
- Learn to write clear prompts. Precise requirements, examples and constraints get better results, much like a good specification does.
- Always verify. Check generated code by compiling and simulating it, and check facts against primary sources.
- Respect confidentiality. Follow your company's rules on what design data may be shared with which tools.
- Move up the value chain. Use the time saved for architecture, review and the judgement calls AI cannot make for you.
FAQ
Who is Dharmesh Shah?
The co-founder and CTO of HubSpot, an angel investor and the writer of the simple.ai newsletter.
What is the main message of "How to Compete with AI and Win"?
Compete by using AI, not by fighting it. The risk is falling behind people who use AI well, not AI itself.
Will AI take engineering jobs?
Shah's view is that AI changes jobs more than it removes them, and that people who use it well will take on bigger and better work. Individual outcomes will vary by role and employer.
How long is the talk?
About 15 minutes.


Your comments will be moderated before it appears here.