How AI Rewrites Work and Power: When One Person Becomes a Company

00:00:00: The issue is people keep thinking that there's a finish line and the reality of it, AI is a marathon.

00:00:07: It's a long race I mean even though a marathon has a finishline.

00:00:10: but we are in the beginning of this race And There Is No Clear Winner.

00:00:18: Hello!

00:00:19: Welcome to Big Bang Tech Report Metvi, Matthew Gallagher... ...and A Story That Just Hits The New York Times.

00:00:28: Metvi shows how AI can turn a one-man company into business with massive scale.

00:00:35: So the big question is, are we looking at wild exception?

00:00:40: Or it's that future of new businesses?

00:00:44: My guest is Alvin Grayland.

00:00:46: Alvin great to have you here!

00:00:48: Hey Hank good to see again.

00:00:49: It has been couple weeks and lot things happening in this space.

00:00:55: Like always, let's start with the big one.

00:00:59: Is this a crazy exception or is it starting in new business areas?

00:01:04: Very small AI-powered companies to future?

00:01:09: We've talked about that maybe two months ago where we said will these new tools enable companies who are very small and become effective?

00:01:20: And clearly I think Matthew Gallagher, his example is a good example of the single person.

00:01:26: later he hired as brother so it became two-person company.

00:01:29: But to be able do what used to take dozens or hundreds people I actually don't think this is an exception, but what might be an exception was the scale of his creation.

00:01:43: In first year he had over four hundred million dollars in revenue.

00:01:47: that's quite impressive for a single person company.

00:01:51: if you look back at its background it has the type skills we talked about before.

00:01:55: you know, he was both an entrepreneur and a coder or technologist.

00:02:01: Right?

00:02:01: And so we said that's the kind of skill set you would need to be able do this type of

00:02:06: company.".

00:02:07: He also found a segment in his business which is relatively low-development work.

00:02:15: So he created the website on the marketplace for JLP One drugs.

00:02:26: So then you're essentially just a channel for distribution and for ordering, which if you have the right tools.

00:02:33: You can make a website and make a commerce site very quickly.

00:02:36: so I think he was fortunate to find the right niche at And he definitely leveraged all the tools out there to make that happen.

00:02:47: So it's not a role model for thousands and thousands of people who will leave university right now, start an entrepreneur career?

00:02:56: I think you could say this is a role-model in starting your business but most people should have expectations about creating a billion dollar company.

00:03:07: I think that's the difference.

00:03:09: I actually agree with what you said earlier, that longer term more and more people will move from big corporations to being in kind of self-employed type a situation which may not be bad thing because it creates lot more autonomy.

00:03:26: but when everybody does that the type of opportunities become saturated or, you know.

00:03:35: The arbitrage opportunities go away very quickly right so...

00:03:39: That's something like what is a startup culture?

00:03:44: Three to four roommates come together have an idea share their ideas build up company and then they are looking for employees.

00:03:52: But this is very different from that, right?

00:03:55: I mean, actually came from the startup world.

00:03:57: I'm actually in Silicon Valley Right now where kind of that's the center and traditional start up.

00:04:02: what happens Is you have an idea You get a few teammates you go on to get DC money you build An idea that solves A problem.

00:04:10: That wasn't that was not solved well before And you do it, five X ten X better than anybody else.

00:04:17: That's kind of the general formula.

00:04:18: and then at some point You keep raising more money and then you go public.

00:04:25: What Matthew Gallagher is trying to what his example shows Is that if there are opportunities To maybe actually disrupt Some of these existing businesses that Are relatively simple Businesses not necessarily developing but distributing or selling, or marketing things that does not require new innovation.

00:04:46: It's maybe an innovation in terms of efficiency but not necessarily a brand-new product.

00:04:53: because to build and design let say you want to build a new silicon chip Or if you wanted to design the new drug it is going take team with people whereas If your just selling something That can be automated And done fairly well With todays tools.

00:05:09: So there's definitely opportunity.

00:05:10: I'm just saying that people need to temper what they believe, the upside of where this potential can go.

00:05:21: so... This is fantastic because it's a real productivity story and i think AI is mostly a productivity story but this one very exceptional.

00:05:35: I think AI is a productivity story today.

00:05:37: But at some point in the near future, AI actually becomes an innovation story as well because we will find that the tools would get good enough to be able to create new molecules and new drugs or solutions but those are going to become one of our biggest models.

00:06:00: You look at something like AlphaFold, everybody thinks that AlphaFolt is creating new protein.

00:06:05: folding must be a giant model.

00:06:06: It's only ninety-three million parameters.

00:06:09: it actually very tiny models can run on my laptop.

00:06:13: So if you have specialized data and specialized people who know how to train them then they create unique capabilities.

00:06:21: so both right?

00:06:24: It could be about efficiency or productivity about innovation when the right people with the right background are leveraging

00:06:34: it.

00:06:34: You're not just watching this trend, you also a co-author of the Stanford paper on enterprise AI.

00:06:43: so let me ask what is your message for that report?

00:06:48: The Enterprise AI playbook came out last week and was co-authored by Lisa Piria and Eric Greenhouse And so it's out of the Stanford Digital Economy Lab.

00:07:01: We've been working on this for about six months, and what really shows is that The majority of problems are keeping people from successfully deploying AI.

00:07:09: It not technology or model.

00:07:12: It's all well now but seventy-eighty percent of it Is organization as people with incentive systems?

00:07:19: Its you know fear its organizational right.

00:07:26: The good news about that is if it's organizational, its something any company can fix.

00:07:32: Anybody with a good leader who sets the right objectives or gets involved and encourages measures properly to work together they will be successful in implementing what you saw.

00:07:47: last year there was a paper out of MIT that said oh you know ninety five percent of AI systems fail.

00:07:52: so this stuff doesn't and we actually found the opposite.

00:07:56: It's not that it doesn't work, Actually works very well right.

00:07:58: in fact even companies have had a very ill managed data that is not well sorted or labeled.

00:08:07: they still got hyperactivity gains.

00:08:10: its really about making sure your workflow integrates with AI versus just inserting AI into existing workflows.

00:08:19: You look at, like the example we just talked about with Matthew where he essentially just figured okay what can I do to use AI to do marketing?

00:08:29: Can AI do customer service and manage my inventory etc.

00:08:36: And when you It's different than somebody saying, okay well I have a sales company.

00:08:40: How can i just replace one piece of it?

00:08:42: If you replace One Piece or One Function then everybody else is the same.

00:08:47: You're not going to get any productivity gains because that one thing might be ten times faster.

00:08:50: But they'll be waiting for all The other things before and after it so you don't see Any real benefit right?

00:08:56: And I think this yeah Yeah.

00:08:59: So I think This Is actually the key is if you're Going To do AI Don't Do it Just For A single function.

00:09:05: read, look at the entire flow of your service or product.

00:09:11: Of your internal operations and find out where all the places that it could benefit.

00:09:16: And then when you streamline that process You automatically get a lot.

00:09:22: But to understand AI as a CEO, you must have the mindset of students.

00:09:29: You need someone who wants to learn and understand what's going on in the future.

00:09:37: Do you see that people go back into their mindsets?

00:09:48: Stanford actually has an executive program, and they have these sometimes one or two week programs where it will bring in very senior executives.

00:09:56: Sometimes CEOs and CXOs to come in and understand what's happening with technology And when they go back then Mindset that you're talking about in terms of understanding, okay it can do these things and not be the resistance.

00:10:13: The thing also was interesting from study is essentially In half of cases models were completely commodity.

00:10:22: It didn't matter which model they picked.

00:10:24: They thought all the same.

00:10:26: It's really about scaffolding how to put together And workflow Once done right than most of the job is done.

00:10:37: There are certain cases, let's say if you're doing coding people do prefer certain models like Claude this very good at coding but for a lot of things lets start looking at let us say invoicing or your'e looking at monitoring cameras.

00:10:51: there's alot of models that could do these types and do it well.

00:10:55: in fact what we also found was as agents came in.

00:11:00: Companies who are implementing agentic technologies were getting an average something like seventy percent increase in productivity, companies that we're doing on high autonomy essentially saying have human and loop it only when something breaks then talk to the human.

00:11:15: you're getting maybe forty-to fifty percent improvement in productivity And there's a lot of humans.

00:11:19: everything has be checked and approved by the human.

00:11:22: You get about twenty percent improvement in productivity.

00:11:25: The more you offload, the more you delegate to AI and hide your productivity.

00:11:30: right now that lesson there is not saying hey I'll upload everything but they're less than their is understand which task when it fails that's reversible And which things are repeated activity.

00:11:46: so if something repeats being done every day or hour then automate right, and let AI make those decisions.

00:11:54: But if it's a decision that you know once you make it ,you can't change it.

00:11:58: then having humans in the loop makes a lot of sense so really being able to differentiate.

00:12:05: but if productivity goes up very fast what happens to people?

00:12:09: What happened to office jobs for example?

00:12:11: we have already discussed this.

00:12:13: but uh...what is your advice to people who feel their job could be automated?

00:12:19: actually one of the findings I find a little disturbing in the sense of The technology that we Of the companies that we had talked to You know, essentially were things That are probably from A year.

00:12:34: just some of them made From two years ago because it takes such a long time To go through large companies of deploying these technologies and all of them have.

00:12:41: they Have at least three months of value before We even consider them to be a successful case to Be put into interview.

00:12:49: Even In that Case forty five percent of The leaders of these companies said we will.

00:12:56: when we see this productivity We will replace the humans with the productivity gains, which means you know That's almost half of the people Half for the company would choose in replacement.

00:13:06: and then another nineteen percent Of them said well.

00:13:09: We won't replace people because that's not nice.

00:13:12: But what we will do is stop hiring Right?

00:13:15: And if you stopped it put that together that's about sixty five percent the total hiring, which means essentially two-thirds of case.

00:13:25: either you're not hiring or your displacing existing workers.

00:13:31: Which has a same impact on the labor force and that's with old technology year to years ago.

00:13:37: just imagine what is happening today how much AI capabilities have improved over even last three months.

00:13:47: I would expect these numbers go up higher over time, and I think that this is where the fear comes from.

00:13:54: And it's warranted.

00:13:56: This also why as we talked about earlier, governments need to start thinking how do we put in programs?

00:14:04: either slow down or create more incentives for employers keep hiring on their workforce?

00:14:14: Or at least provide enough benefits?

00:14:16: so there a soft landing.

00:14:19: So let's make it a little bit more practical because this change will first hit the normal office day.

00:14:26: We have right now, we have meetings emails slides reports follow-ups.

00:14:31: what changes first?

00:14:35: I mean i think its actually going to change anybody that is in front of computer.

00:14:41: if you're doing repetitive work.

00:14:44: there are high exposure to automation And it changes the middle layer of management actually a lot that I think people don't talk about because we keep talking about junior workers.

00:14:55: Yes, there will be less hiring of those unless you're really skilled at AI.

00:15:00: maybe then say hey can take over five of your prior guys and am new person but But the middle layer used to be.

00:15:08: their job was to pass information from the lower layers, too.

00:15:12: The higher layers.

00:15:13: and what's happening right now if you If you put in the right framework for Information transfer with these agents And you put it automatic monitoring systems some of the things that we talked about on our last episode That um?

00:15:29: The CEO is no longer the last to know or maybe your senior manager, it may be not necessary to CEO.

00:15:37: But the senior managers no longer do.

00:15:39: people who know least about organization because they could potentially have a direct line into operational data and need somebody pass again in every filter you lose information.

00:15:52: everybody is hiding something that might benefit them.

00:15:55: so I think A real question and there's a lot of people in that segment Of the job market.

00:16:06: Yeah, so we see less admin more judgment Less execution More responsibility

00:16:14: And That In some ways I think will make not just companies more efficient but Make their society more more efficient.

00:16:23: But you know when they're more efficient The reality is that There will be People who won't Be able to find a place To go Right, so what happens to them?

00:16:31: So this is the government issue that needs to get addressed.

00:16:38: Yeah but do you think we will see... What have had in Germany in the eighties?

00:16:45: four days or thirty-five hours weeks and then we'll see a thirty one hour week, twenty five hours week.

00:16:51: Is it something which can be seen in the future where only work for two day's a week?

00:16:57: I think would be the optimal situation that more companies would say, rather than firing twenty-first percent of my staff I'll move to a four day week workweek.

00:17:07: Rather then finding forty per cent as staff i will move to three days workweek.

00:17:12: Now the question is do you maintain the same pay?

00:17:14: If You Do Then The Shareholders Are Going To Say Well You Just Got All Those Proactivity Gains.

00:17:19: Why Are You Giving Your People More Time Off?

00:17:23: There Will Be Certain and more ethical, moral leaders that would make the choice to say I'll give my people more free time.

00:17:39: That he want that the government wants everybody to move to a four-day workweek.

00:17:43: Maybe then it essentially regulates and enforces that type of working model, Then that might actually make more sense.

00:17:51: if you allow companies to work on their own I think It will be a minority of company's who would make that choice.

00:18:00: Let's switch to our, yeah.

00:18:01: Our question we have nearly in every episode and everyday people ask me Who is building the best machine?

00:18:09: Everybody asked the same question who was ahead?

00:18:12: How do you see Claude quen open my eye?

00:18:15: where are we right now?

00:18:16: um

00:18:18: I think You know that the issue is.

00:18:21: People keep thinking that there Is a finish line And reality AI is a marathon.

00:18:27: It's it's a long race.

00:18:28: I mean even a marathon has the finish line, but we are in the beginning of this race and There's no clear winner right?

00:18:35: In the sense if there Is maybe a leader for the week or leader for today And then may be a leader in a certain segment But not another segments The reality that all these labs are doing very similar things and i think it really doesn't matter over more than a week or month, who is the leader?

00:18:58: Because I feel like right now we're actually taking the wrong approach from an entire industry perspective.

00:19:06: This focus on being ahead is very energy-intensive, compute intensive human effort intensive but it's not necessarily generating value.

00:19:21: And you look at what happened last week with the Cloud Code source leak, right?

00:19:26: Their source code leaked.

00:19:29: What took probably a billion dollars of development is now open sourced to everyone and I would expect that over the next one month probably every company will look at this code and say what are the things i can learn from it probably more important than Claude Opus or Claude Sonnet, the model in terms of what brought value.

00:19:57: And this is also what the research that I was talking about showed us.

00:20:01: how you use AI is more important then which AI you choose and the framework will allow at least a best available frame work now be available to every model maker.

00:20:16: But which model do you use?

00:20:18: And

00:20:19: so I actually, I used a combination of them.

00:20:21: but for coding tasks i still use clod and clod code, clod co-work...I think they're very functional.

00:20:29: For graphics I use Gemini or nano banana.

00:20:37: So from some writing chat GBT.

00:20:43: And if I'm looking at news and tweets, like I go to GROC or to Proplexity so different models provide a different value right?

00:20:54: So again i don't know of anyone.

00:20:57: anybody who is serious uses only one model for everything.

00:21:02: Okay what are the biggest AI mistake leaders make now?

00:21:08: What's the biggest mistake?

00:21:09: you said

00:21:10: Yeah, it's a real mistake.

00:21:12: Do you see something right now?

00:21:13: You're in San Francisco right now Right?

00:21:16: yeah?

00:21:16: and there you are in the middle of everything.

00:21:19: And do you see that something is going wrong?

00:21:23: I mean from me a company implementation perspective or from an industry perspective and lab perspective

00:21:31: From the company from the company that the companies are not adapting its fast enough.

00:21:35: and what?

00:21:36: What do you say there?

00:21:38: The key is really a willingness to fail from the leader, right?

00:21:44: So if you're company leader don't expect that use.

00:21:49: AI is gonna be perfect and everybody's going to have two x improvement in productivity.

00:21:54: And there will no snacks.

00:21:55: The reality is this new technology.

00:21:58: There are lot of resistance because of fear.

00:22:02: we talked about.

00:22:03: Things will fail for multiple types.

00:22:05: The one actually interesting finding that we had from the paper was about sixty-five percent of companies who were successful also failed first.

00:22:18: So they failed first and then kept going, and a hundred per cent people after failure were the same leaders on their projects during the fail project.

00:22:31: so don't punish them when you fail Because if you do, then the other person will make the same mistakes.

00:22:36: But re-learn this lesson again and again.

00:22:39: You need to let people have permission of fail.

00:22:42: And you need to encourage them.

00:22:44: The CEOs have to be in meetings.

00:22:46: they had come.

00:22:47: They show that care.

00:22:49: This is not a pilot project.

00:22:52: That's just for test.

00:22:53: It has been a mandate For change.

00:22:57: When we did it Good things happened.

00:23:00: In fact yesterday I was called With the staff of the Stanford Digital Economy Lab and they had an example of a company that all this The CO did.

00:23:11: was he said in the next year.

00:23:13: I want everybody to double your productivity.

00:23:15: Right, I mean it sounds like a very crazy Crazy mandate but the fact that they did If you, they went and then show the chart when they announced this mandate to next year or actually maybe three months.

00:23:30: Actually over Three Months They essentially productivity per person doubled.

00:23:35: right because he kept every meeting He's like how are using AI?

00:23:39: What do we use in help make yourself better And completely pushing that?

00:23:45: if You Use This Technology Properly It Will Absolutely Bring Benefits.

00:23:51: So AI is not just a changing tool, it's changing the company itself.

00:23:56: It has changed productivity and office work as well as what leaders must do next.

00:24:03: The question is no longer will this happen?

00:24:07: And the real question is who moves first in every industry or not?

00:24:16: I think moving first matters which give you an advantage but isn't sustainable.

00:24:21: because these technologies are changing so quickly that even if you deployed right now with a model and didn't change it, then whoever the new person comes in within another person industry they may actually have to be able to leapfrog.

00:24:40: It's not just doing at first but maintaining a culture of innovation or learning like what you said.

00:24:47: whoever's doing this, you need to keep up-to-date every day because the industry is changing on a daily basis.

00:24:54: Sometimes an hourly basis and if your behind then you will lose.

00:24:58: but what that also does?

00:24:59: it really makes the pace of competition much higher.

00:25:07: Yeah,

00:25:12: today I talked to an entrepreneur and he said okay in the beginning.

00:25:16: I was against AI And I say ok what's that?

00:25:19: They said Ok then i learned a little bit.

00:25:23: Then I thought my productivity went up.

00:25:28: So when I have charged until now for four hours can do it one hour He will charge it for three or two hours and then he has a lot of much more of Benefits.

00:25:41: And so I think that's the story we see right now in Germany.

00:25:46: More and more is this small middle stanza SME?

00:25:49: That people realize that It does no question being against it, it just questions learning and having much more profit

00:25:59: Although also thing that as a temporary that's a temporary arbitrage opportunity of the benefits because as more and more people know this, they will be able to say hey I think you use AI to help.

00:26:13: You only took an hour so maybe not charged for three hours or find another person who would charge about one-and-a half in short term.

00:26:25: yes there are advantages but over time pretty much everybody will start to integrate this into their workflow.

00:26:36: We'll see.

00:26:37: Alvin, thank you very much!

00:26:38: This was the Big Bang Tech Report and I think we have a lot of stories.