Latest / iwon™ podcast with Michael J Thomas / Amazon, Meta and Uber + Founders Building with AI | iwon podcast Summer Sessions with Michael J Thomas | Episode 11

Transcript

Amazon, Meta and Uber + Founders Building with AI | iwon podcast Summer Sessions with Michael J Thomas | Episode 11 · iwon™ podcast with Michael J Thomas

• 0:00
  Welcome to the i1 Summer Sessions.
• 0:02
  It's me, Mikey.
• 0:04
  Hopefully this is a bit like Bob Ross narrating you through tech and what's going on.
• 0:10
  This session is designed to cover the summer period.
• 0:14
  We're all out and about and during our holidays and whatever.
• 0:17
  I thought it would be quite useful to sit down and review what's really going on because everybody's talking about it.
• 0:25
  Normal episodes will resume because after all, it wasn't overnight.
• 0:30
  And I try and bring on the best guests to tell you stories from people who've been there and done it.
• 0:35
  Please do subscribe if you haven't already.
• 0:37
  And please put your feet out and relax.
• 0:40
  I will walk you through what's going on with AI with the big companies.
• 0:45
  Like I said.
• 0:46
  Hopefully you say it sounds like Bob Ross narrating.
• 0:50
  Text her if you did.
• 0:52
  Hope you find this soothing nice and I hope you have a wonderful summer holiday.
• 0:55
  I'll be back in two weeks with another episode of this.
• 0:58
  Normal Episodes will resume just after summer.
• 1:01
  Have a fabulous holiday if you're taking one.
• 1:03
  I thought it would be really cool to look at what Amazon, Uber Meta and Humeta Health could teach founders about AI.
• 1:13
  A founder recently sat down with me one to one and he showed me his new products.
• 1:18
  He had built it using AI in just nine days.
• 1:24
  It works, it looks good and he was really proud of it.
• 1:28
  Then I asked him one question.
• 1:30
  What happens when a paying customer has a problem that you didn't think about?
• 1:36
  What happens when it happens on a weekend, you know, out of hours?
• 1:42
  He didn't have much to say.
• 1:45
  He wasn't entirely sure.
• 1:47
  I mean really, it was just a prototype.
• 1:50
  AI has completely changed what founders can build really quickly.
• 1:54
  In the past you might have needed a technical co founder, a team of developers and months of work.
• 2:00
  But now you can build a first version in just days.
• 2:04
  That's amazing.
• 2:05
  But there is a bit of a problem.
• 2:08
  Getting something to work for a demonstration is not the same as building something robust.
• 2:15
  The good news is that showcasing a concept is much simpler than ever before.
• 2:23
  Almost anyone can do it honestly.
• 2:25
  Hit up level whatever your tool of choice is.
• 2:29
  Base 44, whatever.
• 2:30
  None of them are sponsoring me.
• 2:32
  I'm just throwing some names out there.
• 2:34
  The bad news is that it's still difficult to build something that really is secure, compliant, reliable.
• 2:43
  Most importantly, if you're building a software business, especially an enterprise, being able to provide proper support, possibly more difficult than it's ever been.
• 2:54
  Because having read what I've read, there are also tons more vulnerabilities this is where small businesses especially and bootstrapping founders can get into serious trouble.
• 3:09
  A demonstration could be amazing.
• 3:11
  It can impress investors and attract even your early enterprise customers.
• 3:17
  And it could look completely finished.
• 3:21
  But customers don't buy demonstrations, they buy everything that happens after it.
• 3:26
  That's what they bought.
• 3:29
  That means things like the one off use case, the security, the reliability, the SoC2, all, all that stuff and most importantly, the support needed when things don't work as they should.
• 3:42
  That means, you know, the Sunday morning phone call, the middle of the night call when something important to their business is broken that you provide.
• 3:51
  That's really when you find out whether you built a product or merely built something that works.
• 3:59
  Even Amazon has struggled with this.
• 4:02
  Amazon has some of the best engineers and systems there are.
• 4:07
  Recently, Amazon Web Services CEO, his name is Matt Garman, explored different ways of using AI effectively.
• 4:15
  And he talked about the first being vibe coding.
• 4:18
  And there's a danger that somebody accepts the output as is.
• 4:23
  The second one is augmented coding.
• 4:26
  AI accelerates the work, but there are experienced people in the background who remain responsible for its quality.
• 4:33
  And that distinction sounds simple, but maintaining standards that you need and consistency then that you need across teams of thousands is not easy.
• 4:47
  The FT and the Guardian reported that Amazon's coding agent, which I believe is called Kiro K I R O if you want to look this up, was involved in a interruption out after part of a production environment where
• 5:02
  they were developing was deleted and then recreated.
• 5:07
  This caused 13 hours of disruption.
• 5:10
  Like Amazon, think about the cost to them.
• 5:14
  Amazon did come out and said that AI hadn't caused the interruption.
• 5:21
  It said the problem was caused by misconfigured access controls and disputed the claim that it had affected Amazon Web Services.
• 5:29
  But Amazon also said it had introduced additional safeguards after the event, including mandatory peer review.
• 5:39
  Peer review Humans.
• 5:42
  That's the important part.
• 5:44
  Whatever label you place on it, Amazon still learned from it and strengthened the controls around access to its production environments.
• 5:52
  Amazon has the engineering talent, the review processes and enormous resources.
• 5:59
  So if you're a small 5 or 10 person startup where everybody is racing to ship new features and to build new things, and I'm not saying this is an argument against using AI as an assist, quite the opposite.
• 6:13
  It's an argument for reviewing what the word assisted actually means inside your company.
• 6:19
  AI could write the first pass of the code, but humans still need to understand it, test it, challenge it and ultimately take full responsibility for it.
• 6:29
  That's a lot to ask somebody.
• 6:31
  And that brings me to the second point.
• 6:33
  More does not automatically mean more.
• 6:36
  What do I mean by that?
• 6:37
  Uber offers an interesting lesson on this.
• 6:40
  Essentially, the TLDR version of this would be More code does not automatically mean more value Uber, its development teams adopted cloud code.
• 6:53
  Early usage spiraled, but so did the amount they spent.
• 6:58
  They reportedly used their full year budget for AI in the first four months.
• 7:04
  It's the OO enter.
• 7:05
  McDonald's then questioned whether the increase could be connected directly to more useful features for customers.
• 7:14
  He wasn't sure.
• 7:16
  And that's the bit that founders should pay attention to.
• 7:18
  More work may have been completed technically because more code was being produced, but the commercial value wasn't clear.
• 7:28
  It reminds me of something that I see in marketing all the time.
• 7:32
  Doing more content does not always create more demand.
• 7:36
  More website traffic does not always mean more customers.
• 7:40
  More meetings do not always lead to better choices, better decisions.
• 7:46
  And more code does not always create a better product either.
• 7:50
  Activity, that is what you put in is easy to measure, but the value of that effort is much harder to measure.
• 7:58
  That doesn't make value less important, it makes the measurement more important.
• 8:05
  The biggest AI companies are still working it out.
• 8:08
  Meta is another example.
• 8:10
  They were projected to spend as much as 145 billion on AI infrastructure in this entire year.
• 8:17
  It's one of the largest corporate bets, let alone on AI on anything ever.
• 8:24
  At the same time, they're now exploring ways to commercialize some of that.
• 8:30
  So the data centers.
• 8:32
  Recently Reuters reported that Meta was considering turning it in part of it, certainly into a cloud business that could sell access to spare computing power.
• 8:43
  And they also reported that Meta and Anthropic were having those very discussions about Meta sharing up to $10 billion worth of compute over the next two years.
• 8:56
  Now, those discussions may not result in anything, to be clear, but that doesn't mean Meta has lost faith in AI either.
• 9:05
  It just means Meta starting to think about the infrastructure.
• 9:09
  How will it be used?
• 9:11
  Who else might pay for it?
• 9:14
  How can that become an additional source of revenue for us?
• 9:17
  Even at Meta scale, building capacity is only one part of it.
• 9:23
  There still needs to be a business.
• 9:26
  And so for founders, you should ask yourself the same questions before investing too much time and money into every new AI inverted commerce opportunity.
• 9:37
  Sure, you could build it, but who needs it?
• 9:40
  What would they pay for it?
• 9:41
  Why will they continue using it, and what responsibility?
• 9:47
  Talked about it earlier.
• 9:48
  Support, et cetera.
• 9:50
  Will you be accepting by doing it?
• 9:53
  I get it that those questions are less exciting than generating a working prototype for a new idea in a weekend, but they're also the questions that determine whether you have a business around AI or AI flavored or AI
• 10:12
  assisted that actually lasts.
• 10:15
  And Humana Health had a very similar reckoning recently.
• 10:23
  Their founder is called Dr.
• 10:25
  Jeremy Freese.
• 10:26
  Humada Health.
• 10:28
  The company operates in healthcare, as the name suggests.
• 10:32
  And they've gotta be really careful, right?
• 10:34
  They're dealing with people's personal, medical, private information, right, as well as clinical processes, security, and obviously the serious consequences if those things fell into the wrong hands.
• 10:49
  Writing in Forbes, Freese argued that building an MVP may now appear easier than it's ever been.
• 10:58
  But producing and deploying something reliably at scale is not easier.
• 11:06
  And actually it's also really difficult to audit something that you've not got your hands on.
• 11:14
  The difficult work sits in what you'd call the final mile.
• 11:19
  The final mile includes everything that your demonstration for your new software does not show.
• 11:24
  Security that can survive a real attack.
• 11:26
  Compliant software that holds up during an audit, integrations that keep working as other systems change because we don't operate on an island anymore.
• 11:39
  And that last 20%, that final mile may take far longer than the first 80%.
• 11:45
  Producing something sexy, making people want, may look impressive on LinkedIn to say that we've got this new feature or whatever, but what if it doesn't work the way you expect it to?
• 11:57
  And that's where the actual product lives.
• 12:01
  And that's why build versus buy is is totally changing.
• 12:07
  Companies used to ask, especially in enterprise, can we build this ourselves?
• 12:13
  AI has made it easier for them to say yes.
• 12:16
  But as with everything, that's not the full story.
• 12:21
  The better question is, does our future competitive advantage live inside this?
• 12:28
  Because on one hand, being able to build something doesn't mean you should own it forever.
• 12:35
  Every piece of software creates an ongoing responsibility.
• 12:39
  What I mean by that is all the tedious stuff, right?
• 12:41
  The maintenance, the updates, the security reviews, the documentation, the support, the infrastructure, the compliance, the internal technical knowledge.
• 12:51
  Somebody still needs to assure customers that the right quality and security is in place.
• 12:58
  AI may have made the first version of something a piece of cake, but it has made it even more important to keep on top of everything that comes after that.
• 13:09
  And Gartner did some research and they predicted that more than 40% of agentic AI projects will be canned by the end of 2027 because it costs more to make, they're not sure about the value, and there are very
• 13:28
  few risk controls in place.
• 13:29
  There is an optimistic side to this.
• 13:31
  AI has raised the floor, I.
• 13:33
  E.
• 13:33
  It's made it easier to make Prototypes.
• 13:36
  But it also makes the ceiling, I.
• 13:39
  E.
• 13:39
  Doing things the right way much, much more valuable when every demo looks polished.
• 13:46
  Depth, knowledge, security, compliance, all those things become a real differentiator.
• 13:53
  So when everybody claims to use AI, better trust becomes the key differentiator.
• 13:58
  What companies claim versus what they deliver is absolutely everything.
• 14:04
  So obsess like crazy founders over security, support, compliance and reliability that that can become your moat.
• 14:14
  That is also the work that nobody gets excited about.
• 14:18
  The work buyers rarely ask a million things about until they realize it's not there.
• 14:25
  It's the work that allows both you and your customers to sleep at night.
• 14:31
  Every high profile failure teaches buyers too.
• 14:37
  It teaches them to ask better questions about what's behind it all, such as who's responsible when things go wrong?
• 14:45
  Who's going to support me?
• 14:47
  How is this tested?
• 14:49
  What information can IT access?
• 14:53
  Can we audit the decision making part?
• 14:58
  Is everything safe?
• 15:00
  And of course the question that pains every startup versus every huge company, will this company still be there when we need help?
• 15:11
  So those questions very much favor founders who have built something real and deep and proper.
• 15:19
  So three questions is one.
• 15:22
  What would happen if a customer stress tested your product tomorrow?
• 15:27
  Which things are you not entirely sure about?
• 15:30
  How are you keeping on top of everything?
• 15:32
  That means the small details, when you can knock out code in 30 seconds, how are you keeping on top of that and making sure that the back end, the security, the compliance, the support, everything's in place?
• 15:45
  Is the depth, I.
• 15:48
  E.
• 15:49
  How much effort you've gone to to build everything properly clear and the value of your product clear to buyers?
• 15:57
  In a world where another competitor could create something around the corner, are you demonstrating enough real value to make customers want to stay with you?
• 16:08
  Building is becoming so much easier, but responsibility hasn't changed.
• 16:14
  I use AI.
• 16:15
  I think it's one of the most important technological shifts ever.
• 16:21
  But believing in AI does not mean switching off our own judgment.
• 16:25
  Because building with AI, whilst easy, can leave us vulnerable.
• 16:31
  Building a business people can depend on is still difficult and that's exactly where the opportunities lie.
• 16:37
  I'd love to hear your thoughts.
• 16:39
  What are you seeing?
• 16:40
  What do you think?
• 16:41
  Let me know in the comments or pop me an email.
• 16:46
  I'm Michael and this podcast is for founders like you.
• 16:50
  Please do like comment, subscribe and give us ratings.
• 16:53
  All of that stuff helps.
• 16:56
  Have a great day and thank you for listening.