TBPN Sep 21, 2026 24m 14m saved
With Vinod Khosla, founder of Khosla Ventures
Almost nobody would hand a software agent legal authority to act for them today. Vinod Khosla's estimate is that 40% or 50% of people will do it by 2030.
The standard argument about personal agents is that whoever holds your calendar, your email and your messages holds you. Khosla thinks that lock-in evaporates, because the agent itself will do the offboarding, and that what is left is something far harder to accumulate.
"I believe the moat will be trust."
Khosla founded Khosla Ventures and co-founded Sun Microsystems, and says he first made the case for a consumer-side AI about four years ago, on the argument that the largest opening was protecting people from everybody else's AI.
The full interview is covered here so you can skip it. 24 minutes of audio, 10 minutes of reading.
Here are the 7 predictions that matter.
Key Takeaways
Nobody wins personal agents outright — every large lab and platform will ship one, with different trade-offs, and "agents from startups will move faster"
The split that matters is advertising versus everything else, because an ad-funded agent has a second customer
ChatGPT killed the idea that consumers will not pay, and an agent that saves money is the version consumers pay for
Offboarding is not a moat, because a competent agent does the migration for you He points at the prompt-export truce between OpenAI and Anthropic as the precedent
Trust is the moat, and completion is the second one — consumers stay with whatever finishes the job
A human in the loop is a small company's advantage, because a product serving a billion users cannot staff it
The agent should not be an app. Email, text and Slack are where people already are, and that is where it has to work
40% or 50% of people will give an agent power of attorney by 2030, up from close to nobody now
1. No Winner Takes This
The hosts opened by reading a widely shared argument from Alex Kerr: that Silicon Valley is over-optimizing personal agents for outcomes, that people want to browse and daydream rather than delegate, that most lives contain fewer delegable recurring tasks than the industry assumes, and that personal agents will therefore be absorbed into existing products rather than becoming a category.
Khosla disagreed on the category question first.
A broad, diverse market, not a single winner
Well, it's a new category. It'll be a broad category. I do think it'll be very diverse, and I don't think there will be a winner-take-all.
Vinod Khosla
Every large lab and platform will want its own agent, he said, and each will make a different trade-off. On the browsing objection, his answer was that the agent doing the looking and coming back with options and questions is a better version of browsing than doing it yourself, and that Kerr has not used agents enough to see the range.
The trade-off that decides the shape of each product
I think the fundamental difference will be, do you want to be an advertising product or be something else? Do you want to trust matter with your data?
Vinod Khosla
Some people will, he said, and Meta will be successful. Others will want something else.
Startups move faster than platforms here
And I do think agents from startups will move faster than agents from the big companies.
Vinod Khosla
2. The Consumer's Own AI
Asked about the fight between agents and the platforms they shop on, Khosla separated two things that look alike.
Last year's browser wars, he said, were not driven by consumers wanting to leave Chrome. They were driven by companies not wanting Google to control what happens inside the browser. That is a corporate need dressed as a product.
What consumers face is different, and in his telling it is an arms race they are currently losing: a marketing assault run by AI, tuning colors to raise clicks, optimizing feeds, pricing reviews.
The largest opening is an AI that fights the other AIs on the user's behalf
So representing the consumer and having an AI that loyal to the consumer not to a particular company, a particular product, a particular point of view is going to be very, very important.
Vinod Khosla
He said he first made this argument about four years ago. His example of the damage was product reviews: a consumer can no longer read one without having to work out who paid for it.
3. Consumers Will Pay to Save
One host put the standard Silicon Valley position to him — consumers pay to be entertained, not to be productive, so productivity tools end up sold to enterprises. Both hosts noted that Muse is already being used to cancel unused subscriptions and claim refunds, which looks more like buying tax software than scrolling a feed.
ChatGPT settled the question
Google always assumed consumers won't pay. And that was the conventional wisdom and ChatGPT, in my view, destroyed that. For enough value, consumers will pay.
Vinod Khosla
Saving money is the version they pay for
Now, saving you money is really, really valuable for most consumers. And they will pay if something does that.
Vinod Khosla
Travel, shopping and subscriptions all have money sitting in them, he said, and he included himself in the problem.
Even he does not know what he is subscribed to
I have way more subscriptions than I know about.
Vinod Khosla
What makes it work, in his framing, is that the agent is acting as the consumer's own assistant rather than as a company's.
4. Offboarding Kills the Moat
The thesis the hosts put to him was switching cost: once an agent has your calendar, email and messages, you will not move for a marginally better product.
A good agent will move you itself
I think offboarding will be possible, a little bit clumsy, but then if you have a good agent, you tell it to do the offboarding.
Vinod Khosla
And the onboarding onto the next one, he added. His precedent was the moment OpenAI and Anthropic each published instructions for dumping your memories into the other product, fighting over portability because portability was already possible.
Complexity stops being a barrier once the agent handles complexity
AI is just going to be good at doing all these complex things.
Vinod Khosla
His illustration was the effort it currently takes to work out what Google knows about where you were at five o'clock on a given afternoon, and then to turn it off. It is not impossible; it is tedious. Tedium is exactly what an agent removes.
5. Trust, Then Completion
Asked directly what the moat is, then, Khosla named two things.
Trust is the first one, and it is not evenly distributed
I believe the moat will be trust. In a company that develops trust with consumers, and that's where META has a huge disadvantage. So the trust will be the moat.
Vinod Khosla
Completion is the second
I also think completion is going to be really important. Consumers will be loyal to products that get their shit done.
Vinod Khosla
A host offered the mechanism: the agents worth using know when they are stuck. Most agents either report being stuck or simply sit there, while Wajo, the Khosla-backed agent, pulls a human in as a tool and uses the person to finish the job.
That, Khosla said, is hard to copy at scale, and the difficulty runs the opposite way to the usual advantage.
Humans in the loop do not scale to a billion users
So there's an advantage to being small and agile.
Vinod Khosla
For someone who needs the task actually done, he added, paying a little for the human in the loop is much better than not getting it done.
6. The Agent Is Not an App
Khosla's next point was about form factor, and he framed it against the browser discussion.
People live in email, text and Slack, so the agent has to
The right agent won't be an app. It just, where do most people live? They live in their email. They live in text messaging. They live in Slack. They live in other channels.
Vinod Khosla
If there is an app at all, he said, it runs in the background. Sometimes you text the agent; sometimes you pick up the phone; sometimes the agent picks up the phone and talks to a merchant.
Channel constraints are a product failure
So, meeting a consumer where they are, the consumer experience is really, really important, not get stuck in, we do text only, we do browser only, you have to be in our app.
Vinod Khosla
A host agreed from the other end: nobody wants another app on the phone, and the app store is already all AI apps.
On whether an agent should identify itself, Khosla answered with something that had happened to him the day before. He was ordering from Petco, told the agent to pretend to be him, and the agent refused.
The agent works for you and is not you
The agent is your agent, but it's not you.
Vinod Khosla
Transparency of that kind is what builds the trust he had just called the moat, he said, and it has to hold for both sides of the conversation.
7. Power of Attorney by 2030
The last question was whether users will eventually give agents something like power of attorney — the point at which a service says it needs the customer personally on the phone, and the assistant can go no further.
Khosla's answer routed back through the moat and then gave a date.
Almost nobody today, and roughly half within four years
Now, in four years, will that start to happen? I suspect 40% or 50% of people will give a power of attorney to the agent by 2030 because they've had enough interaction and developed enough trust.
Vinod Khosla
Which is the case for startups rather than incumbents
So trust becomes the key issue, and this is why startups may have an advantage. Startups with great transparency will have an advantage over people who already have a brand of lack of trust.
Vinod Khosla
Bonus Insights
Payments are the differentiator he is watching
If you can be absolutely secure and give your agent, your credit card, that's a big deal. And there will be lots of abuses or mistakes in that domain.
Vinod Khosla
He said he is comfortable handing his own card to the agent his firm backs, because of protections he expects to be patented and would not describe, and that he can be certain the card is not reused for anything he did not authorize. A host's gloss was that spending limits belong in deterministic code rather than in a model that can mistake $100 for $100,000.
He invested in Wajo on day one and helped shape the idea
Well, we invested in it day one. In fact, we brainstormed with Shivani, who's just a great technologist and a real understanding of the human factors.
Vinod Khosla
Empathy for the consumer, he said, is the thing he was looking for, and it is what he means by the right mindset for this category.
Access opens at the end of the month
Today, it is not free access. It's limited access by invite only.
Vinod Khosla
By September 30th, it'll be a free relationship. So by the end of the month, you'll see it pretty open to everybody.
Vinod Khosla
A working example of the delegation problem, from the host's own week
One host described wanting a specialty dog collar after his puppy's surgery: six websites, a set of stores to call, and one question that actually mattered, which was same-day delivery. That is the shape of task he wants handed off, including the payment at the end.
Newfoundlands
Asked what kind of dog he keeps, Khosla named the breed twice and said one of his two reached eleven. A host said his own two, both around 200 pounds, died last year.
Khosla's bottom line is that personal agents will not be won by whoever accumulates the most user data, because the data moves; they will be won by whoever a consumer is willing to let act on their behalf, and that is a slower thing to build and a faster thing to lose.
Products, Companies & Tools Mentioned
Khosla Ventures (Khosla's firm, an investor in Wajo from day one)
Wajo (The personal agent he backs, built by Shivani Poddar; pulls a human in as a tool when it gets stuck, invite-only until September 30)
Meta (Named as the company starting this race with a trust disadvantage, and as the example of an AI tuned to make you click)
ChatGPT (His evidence that consumers will pay for software at scale, against Google's long-held assumption)
OpenAI and Anthropic (The memory-export fight he uses as proof that switching between AI products is already practical)
Google Chrome (The browser wars, which he says were about companies not wanting Google in control rather than about consumers wanting to move)
Petco (Where he was ordering when his agent refused to impersonate him)
Slack (One of the channels he says an agent has to operate inside, alongside email and text)
Listen to the full episode
🔴 YouTube | 🔗 Episode page
Watch the full episode:
More Vinod Khosla interviews and podcast appearances
If this was worth your time, send it to someone closer to the industry than you are.
Get the latest market chatter as it happens: