> "殺人放火金腰帶，修橋補路冇屍骸" — "those who commit crimes wear gold belts, while those who build bridges leave no trace."
The invisible figures doing the real work don't get remembered because it's not sexy. Especially in this self-promoting culture.
This tracks with your point about IT teams vs AI startups. The ones recalibrating gradients and writing manifestos get the $50B valuations. The internal teams actually rebuilding workflows? They're just... IT.
Maybe this is why China's tech ecosystem thrives differently — less emphasis on personal brand heroics, more focus on collective contribution. The cathedral gets built either way, but the culture decides who gets credit for laying the bricks.
I'm not sure if this is a good thing or a bad thing, but it does seem possible that there will at least be some period where "new" IT teams do get celebrated some. Or like, a phase where the hot new job is "AI IT team," that builds all the internal agents and stuff. (There have been a few posts from internal data agent teams that have gotten a lot of attention, so stuff like that).
On the one hand, maybe that's closer to celebrating the bridge builders. On the other hand, it is a new class of bridge builders that are kind of like AI-pilled founders talking about recalibrating gradients, so maybe it's still more of the same.
Could be - something tells me that it would probably just be a short term flux. In longer run it would still swing back to some founder bluffing new things to get investors money.
actually, my own opinion, and take that with a grain of salt because I am observing from afar, is that a lot of these layoffs are just getting rid of people who were not doing much anyway and then using AI as the excuse. Rather like how 80% of staff left Twitter after Elon took over and nothing fell apart.
Exactly. Clearly AI is being used as a scapegoat for layoffs. Why would AI productivity enhancements lead to layoffs? That makes no sense. You can do even more if you have more people. I've never been in a team or company that didn't wish they had more people (but limited by budget).
- Laying people off because of financial difficulties = stock goes down.
- Telling the world it's because of AI productivity enhancements = stock goes up.
Duh. It's an easy spin. It's all theatre.
But also gotta pay for those tokens, so maybe a few people gotta go...
Ultimately, it's just like any other tool: some people will spend too much money doing the wrong things, and others will do the right things and keep their spending under control. We've all seen outrageously expensive SaaS contracts that bring very little value, and reasonably priced contracts that bring a lot of value. AI is no different.
Correct Marco, in the current environment laying off people and proclaiming productivity improvement from AI is just theatre to protect the stock price. That is my theory too but you said it out loud.
Most companies who sell to the consumer market would make a much higher return on investment just by doing data warehousing properly.
Indeed now it is possible to put dimensional models as views over complex operational data models mass market companies can do BI style analytics over just their operational system data using a database as cheap as sql server standard edition. Much cheaper than all those tokens.
I am still of the opinion that the most reliable source of new profit for any mass market company over usd100M revenue is to give the smartest business analysts in the company a copy of Ralph Kimballs Meta5 and access to data presented as a dimensional model.
Much cheaper and much faster than any AI project. I have been saying that for 35 years now. I dont think my opinion will change before I retire.
It depends. I think the big unlock from working with AI is twofold:
- working with raw text / unstructured data, and handling ambiguity. Computers have never been able to do this before. This is a net-new feature. It unlocks countless use cases without the need for analysis. Like "here are our customer complaints, tell me what makes them unhappy".
- speed & automation. New or junior employees can ramp up much faster and do much more by leveraging AI, which can search & read much faster than they can. And senior people who know what they are doing can use AI to automate much of their work or create new tooling (even replacing many of their SaaS vendors entirely).
When used correctly, it's a game changer. When use well, it's an enhancer. And when used incorrectly, it's a hole in your wallet with very little to show for it.
Give a knife to the right person, and they will start a restaurant. Give a knife to the wrong person, and people will die.
well Bill Inmon has been working on Textual Analytics for many years. I have done his training on his product. So I know his product. I would like to sell it but it's a bit pricey.
Your example of:
"Like "here are our customer complaints, tell me what makes them unhappy"."
Bills product has been able to do that for at least 6 years I know about now.
In fact, going right back to the early 90s one of the things we did was call outs to customers who left companies to ask them why they left. One of the issues is that very often they don't tell you the real reason.
One of the key features of Bills softwas was that it could take transcriptions of called into the customers call center and then build data models of what the transcripts said to then be searchable to see if there were common threads as to what people were saying.
But let me assure you customer feedback has been very closely monitored and analysed by many of my customers for more than 30 years. Sure, AIs can read text better now, but it's only a marginal improvement.
Yes, I see AI as being able to on board new employees, especially in training courses. For example? We used to do class room training. Now we do video based training, that has been quite a leap. I just spent more than 350 hours migrating class room training for video training for some software. And yes an AI can keep a close eye on new starters in what they do.
But this too is nothing new and is only incrementally better.
One of the things we have been doing since 1994, something we discovered by complete accident, is this.
We look for high performers in what ever area of the business, usually sales. We then go through their sales to see what they are doing. And then we interview them. Then we see what can be replicated across the sales force.
I will share with you the story about how we came to do this. In 1994 I was working for the Mutual Life Company in Australia as a consultant. They were my IBM Customer and I had resigned from IBM and was now directly working with them.
A gentleman walking into our work area and asked of a man named Ross was there. I said no, that he had moved on and was no longer in our department, I asked the man his name and if I could help him. He was the number one agent in the company. He asked to please talk to me in a meeting room.
He explained that Ross has been sending him prospective customers in spreadsheets and that he had been following up these prospective customers. This prospective customers were how he had been doing so well. Only problem was this violated the companies policies.
So I took this man to the business manager I reported to and asked him to please repeat his story. I thought he was about to get into trouble. The business manager almost exploded with delight. He took him straight to the state manager and he told the story again. And then the state manager took him to the national sales manager and he told his story again.
The national sales manager looked at my business manager and said: "Isn't this exactly the sort of thing that project you have been bugging me with is all about?"
And he said yes. And the national sales manager said, you have your money, make sure you get this man involved, (meaning the number one rep in the country) and work as quickly as you can.
As it happened the business manager I was working for had been trying to get a project approved to introduce a more sophisticated CRM system for the agents. The budget was about $A1M because we had 1,500 agents selling insurance and the national manager was just not that convinced that CRM at the agent level was worth the investment.
But now he had the number one agent in the country who had implemented his own CRM style system in his office and he had been fed leads by this man called Ross for nearly three years and that was a very large part of his success. That these leads were violating company policy was completely overlooked and the CRM project was rolled out to a great deal of success.
Now, would AI help? Yes. But it's marginal how much it would help compared to making such a decision as this one in the first place.
In all my clients where the project has been about "sell more stuff" we have done much more sophisticated analysis of how the most successful people are the most successful and then figured that out and cross trained the rest of the staff. This drives very good return on investment. AI over the top would be marginal at best. We already know how to do these things without AI.
As you say, AI is one more tool. But it is coming on on top of all the other tools we have had for improving business performance and they have been very effective for a very long time.
I think this is probably true, that companies are doing this for a lot of reasons unrelated to AI, or at least, folding those corrections into one big AI one for the optics.
That said, I do also think that some of these companies (block for sure, Coinbase probably, maybe cloudflare) are very genuine in their intent to redesign the company around AI. It might not work, or course, but I'd be very surprised if those places don't try some real experiments along the way.
Brilliant essay! What is your view on the future regarding the companies that are now deeply merging with one or the other AI giant, not just using their models but weaving their workflows so tightly with the AI giants' tools that you can barely untangle them anymore. And they are doing this for speed and competitive advantage. What does their future look like? What do you think?
That seems rather unwise given how easy it is to switch between models (unlike switching databases, etc.). But vendor lock-in is nothing new. Some people are tempted by golden handcuffs.
I think it's probably kinda fine? That's more or less how companies were with cloud providers. They built on AWS or Azure, and maybe, eventually moved to being multi-cloud, but a lot started with their apps being really intertwined with that provider's specific offerings. AI models aren't quite the same as a rented computer running the same open source software, but it seems like they're converging a good bit. Plus, the trend now is that the harness matters as much as the model. And if you can build a better harness by leaning into the model's (or model provider's) particular quirks, that might end up being a better product than one that can constantly respond to the new model updates.
I would start my own company but I dont want to have to layoff half my company (50k people) due to AI productivity gains so I will remain a worker been and save jobs and the economy.
There's a Chinese saying:
> "殺人放火金腰帶，修橋補路冇屍骸" — "those who commit crimes wear gold belts, while those who build bridges leave no trace."
The invisible figures doing the real work don't get remembered because it's not sexy. Especially in this self-promoting culture.
This tracks with your point about IT teams vs AI startups. The ones recalibrating gradients and writing manifestos get the $50B valuations. The internal teams actually rebuilding workflows? They're just... IT.
Maybe this is why China's tech ecosystem thrives differently — less emphasis on personal brand heroics, more focus on collective contribution. The cathedral gets built either way, but the culture decides who gets credit for laying the bricks.
I'm not sure if this is a good thing or a bad thing, but it does seem possible that there will at least be some period where "new" IT teams do get celebrated some. Or like, a phase where the hot new job is "AI IT team," that builds all the internal agents and stuff. (There have been a few posts from internal data agent teams that have gotten a lot of attention, so stuff like that).
On the one hand, maybe that's closer to celebrating the bridge builders. On the other hand, it is a new class of bridge builders that are kind of like AI-pilled founders talking about recalibrating gradients, so maybe it's still more of the same.
Could be - something tells me that it would probably just be a short term flux. In longer run it would still swing back to some founder bluffing new things to get investors money.
Fair, those IT have to be sold something...
Hi Benn,
actually, my own opinion, and take that with a grain of salt because I am observing from afar, is that a lot of these layoffs are just getting rid of people who were not doing much anyway and then using AI as the excuse. Rather like how 80% of staff left Twitter after Elon took over and nothing fell apart.
Exactly. Clearly AI is being used as a scapegoat for layoffs. Why would AI productivity enhancements lead to layoffs? That makes no sense. You can do even more if you have more people. I've never been in a team or company that didn't wish they had more people (but limited by budget).
- Laying people off because of financial difficulties = stock goes down.
- Telling the world it's because of AI productivity enhancements = stock goes up.
Duh. It's an easy spin. It's all theatre.
But also gotta pay for those tokens, so maybe a few people gotta go...
Ultimately, it's just like any other tool: some people will spend too much money doing the wrong things, and others will do the right things and keep their spending under control. We've all seen outrageously expensive SaaS contracts that bring very little value, and reasonably priced contracts that bring a lot of value. AI is no different.
Correct Marco, in the current environment laying off people and proclaiming productivity improvement from AI is just theatre to protect the stock price. That is my theory too but you said it out loud.
Most companies who sell to the consumer market would make a much higher return on investment just by doing data warehousing properly.
Indeed now it is possible to put dimensional models as views over complex operational data models mass market companies can do BI style analytics over just their operational system data using a database as cheap as sql server standard edition. Much cheaper than all those tokens.
I am still of the opinion that the most reliable source of new profit for any mass market company over usd100M revenue is to give the smartest business analysts in the company a copy of Ralph Kimballs Meta5 and access to data presented as a dimensional model.
Much cheaper and much faster than any AI project. I have been saying that for 35 years now. I dont think my opinion will change before I retire.
It depends. I think the big unlock from working with AI is twofold:
- working with raw text / unstructured data, and handling ambiguity. Computers have never been able to do this before. This is a net-new feature. It unlocks countless use cases without the need for analysis. Like "here are our customer complaints, tell me what makes them unhappy".
- speed & automation. New or junior employees can ramp up much faster and do much more by leveraging AI, which can search & read much faster than they can. And senior people who know what they are doing can use AI to automate much of their work or create new tooling (even replacing many of their SaaS vendors entirely).
When used correctly, it's a game changer. When use well, it's an enhancer. And when used incorrectly, it's a hole in your wallet with very little to show for it.
Give a knife to the right person, and they will start a restaurant. Give a knife to the wrong person, and people will die.
Hi Marco,
well Bill Inmon has been working on Textual Analytics for many years. I have done his training on his product. So I know his product. I would like to sell it but it's a bit pricey.
Your example of:
"Like "here are our customer complaints, tell me what makes them unhappy"."
Bills product has been able to do that for at least 6 years I know about now.
In fact, going right back to the early 90s one of the things we did was call outs to customers who left companies to ask them why they left. One of the issues is that very often they don't tell you the real reason.
One of the key features of Bills softwas was that it could take transcriptions of called into the customers call center and then build data models of what the transcripts said to then be searchable to see if there were common threads as to what people were saying.
But let me assure you customer feedback has been very closely monitored and analysed by many of my customers for more than 30 years. Sure, AIs can read text better now, but it's only a marginal improvement.
Yes, I see AI as being able to on board new employees, especially in training courses. For example? We used to do class room training. Now we do video based training, that has been quite a leap. I just spent more than 350 hours migrating class room training for video training for some software. And yes an AI can keep a close eye on new starters in what they do.
But this too is nothing new and is only incrementally better.
One of the things we have been doing since 1994, something we discovered by complete accident, is this.
We look for high performers in what ever area of the business, usually sales. We then go through their sales to see what they are doing. And then we interview them. Then we see what can be replicated across the sales force.
I will share with you the story about how we came to do this. In 1994 I was working for the Mutual Life Company in Australia as a consultant. They were my IBM Customer and I had resigned from IBM and was now directly working with them.
A gentleman walking into our work area and asked of a man named Ross was there. I said no, that he had moved on and was no longer in our department, I asked the man his name and if I could help him. He was the number one agent in the company. He asked to please talk to me in a meeting room.
He explained that Ross has been sending him prospective customers in spreadsheets and that he had been following up these prospective customers. This prospective customers were how he had been doing so well. Only problem was this violated the companies policies.
So I took this man to the business manager I reported to and asked him to please repeat his story. I thought he was about to get into trouble. The business manager almost exploded with delight. He took him straight to the state manager and he told the story again. And then the state manager took him to the national sales manager and he told his story again.
The national sales manager looked at my business manager and said: "Isn't this exactly the sort of thing that project you have been bugging me with is all about?"
And he said yes. And the national sales manager said, you have your money, make sure you get this man involved, (meaning the number one rep in the country) and work as quickly as you can.
As it happened the business manager I was working for had been trying to get a project approved to introduce a more sophisticated CRM system for the agents. The budget was about $A1M because we had 1,500 agents selling insurance and the national manager was just not that convinced that CRM at the agent level was worth the investment.
But now he had the number one agent in the country who had implemented his own CRM style system in his office and he had been fed leads by this man called Ross for nearly three years and that was a very large part of his success. That these leads were violating company policy was completely overlooked and the CRM project was rolled out to a great deal of success.
Now, would AI help? Yes. But it's marginal how much it would help compared to making such a decision as this one in the first place.
In all my clients where the project has been about "sell more stuff" we have done much more sophisticated analysis of how the most successful people are the most successful and then figured that out and cross trained the rest of the staff. This drives very good return on investment. AI over the top would be marginal at best. We already know how to do these things without AI.
As you say, AI is one more tool. But it is coming on on top of all the other tools we have had for improving business performance and they have been very effective for a very long time.
I think this is probably true, that companies are doing this for a lot of reasons unrelated to AI, or at least, folding those corrections into one big AI one for the optics.
That said, I do also think that some of these companies (block for sure, Coinbase probably, maybe cloudflare) are very genuine in their intent to redesign the company around AI. It might not work, or course, but I'd be very surprised if those places don't try some real experiments along the way.
Brilliant essay! What is your view on the future regarding the companies that are now deeply merging with one or the other AI giant, not just using their models but weaving their workflows so tightly with the AI giants' tools that you can barely untangle them anymore. And they are doing this for speed and competitive advantage. What does their future look like? What do you think?
That seems rather unwise given how easy it is to switch between models (unlike switching databases, etc.). But vendor lock-in is nothing new. Some people are tempted by golden handcuffs.
I think it's probably kinda fine? That's more or less how companies were with cloud providers. They built on AWS or Azure, and maybe, eventually moved to being multi-cloud, but a lot started with their apps being really intertwined with that provider's specific offerings. AI models aren't quite the same as a rented computer running the same open source software, but it seems like they're converging a good bit. Plus, the trend now is that the harness matters as much as the model. And if you can build a better harness by leaning into the model's (or model provider's) particular quirks, that might end up being a better product than one that can constantly respond to the new model updates.
I would start my own company but I dont want to have to layoff half my company (50k people) due to AI productivity gains so I will remain a worker been and save jobs and the economy.
not the hero we deserve, but the hero we need