Dear Lewis, my CEO wants AI to do it all. How do I argue for humans?
lewislin.substack.com
lewislin.substack.com
Just want to highlight one element in particular that jumped out at me:
"AI has nothing personal at stake. It doesn't feel the pressure of missing quota or the exhilaration of exceeding it. It doesn't have a mortgage payment riding on that commission check."
So the AI doesn't manipulate power dynamics to control its employees... and that's a bad thing? Okay. (It isn't true anyway; AI can easily do that.)
This has scary implications if extrapolated out to extremes, because a sufficiently advanced AI would do exactly what you’re describing, manipulate power dynamics to get to a superficial outcome.
I get the impression, when talking to conversational AIs, that they're more tuned to convince you that you're right — sycophantry likely minimizes how often people press the RLHF thumbs-down button, and thereby appears more-often-than-warranted in the RLHF fine-tune dataset.
Glossing over the whole AI thing... Maybe we shouldn't be structuring our systems so that the humans are one bad quarter away from financial ruin either
It doesn't meet the objective definition of capitalism, and its quite trite blaming failures along propaganda lines.
Capitalism can't technically exist without a stable store of value, you have to be able to make a profit in purchasing power and that simply isn't possible under certain systems.
Free-markets also can't exist under money-printing for the exact same reason. There are parasites that drive everyone else out of business and then collapse during the final stage of ponzi. Sieving all assets into few hands that can then be seized by government is communism.
What you pretend to be capitalism is in fact socialism, and like most socialist issues these systems create the problems, claim its something else, and then put forth solutions that are not solutions but enable greater control towards totalism. In other words, shock doctrine.
That path leads to extinction while trying to drag everyone else along for the ride.
Its quite evil, and its inevitable when you have those elements present and you can see that if you think along rational principles based in external measure/reality. Most civilizations never make it beyond a certain point.
You don't stay alive by ignoring reality.
Seems like a more reasonable path to me; more logic and less bullshit at the core, keep human creativity.
The director from the Travelers series, basically.
Just consider the potential savings...
But the role of steering a company is needed, that's why I think it's perfect for AI. Developers and VCs write the instructions together and AI runs the company.
Wasn't that exactly my question?
You can just run the codebase through an LLM to cleanse it of any IP entanglements. Open source it under a pseudonym if you're worried about retribution. Whatever parts of the business that doesn't cover... well those are the people you need to hire from the old one.
> First AI came for the artists and I said nothing because I was not an artist...
-- VC's and CEO's while their ship sinks
Drop the CEO and keep the developers instead!
Better yet: consider replacing the shareholders with AI.
But no, no one considers that, because those are the people who have the power. And "replacing with AI" is all about power.
- Do you want an LLM salesbot to close a deal your company isn't actually able to fulfill?
- Do you think your company will use AI more intelligently and reliably than the people who made a popular LLM coding system?
I want my AI support rep to create tasks to engage my team, with all the relevant data linked to it. It should be able to automatically schedule things and ask a human for confirmation.
It should be able to elevate communication to a human in the loop, using whatever mediums of communication makes sense given staff availability and workload.
At no point is it allowed to answer any questions unless the answers are constrained and probably directly quoted from a cited and linked section in our documentation.
In general, it should never confirm or deny things, it should never try to close things or acknowledge the content of any of the user's communication other than to call tools and surface public information which might be relevant to their request.
Most of this is a software architecture problem. The LLM is just there to provide an intuitive and extremely powerful natural language interface for search and tool calling. A little bit of glue between different systems, both internal and external.
If you were an end user of such a system, would you be happy?
It's an incremental stop along the way to truly reliable agentic systems which we can trust with important things.
I suspect Cursor told themselves that they are super-smart AI experts who would never make an amateur mistake like the airline, they will use prompt engineering + RAG. With this, it will be unpossible that the LLM could make a mistake.
As a junior, I often wonder how many deals are signed in exclusive country clubs, on golf courses, and at the dining table with endlessly flowing Maotai.
For a successful career, is it better for one to prioritize network over skills? It seems to me that the latter can be commoditized by AI, while the former can not. Rather than learning Lisp, maybe it's time to pick up golf. I'm only half joking.
Virtually none in our business. (Databases.) What does get deals is listening carefully to what customers actually want and putting together offerings that get it to them at a reasonable price. Incidentally, good sales people are vastly better at this than devs. There are a number of skills that go into it but being a good listener is the most important.
More often that one would like, enterprise SaaS sales isn't about having the best product - it's about convincing the CTO he's going to feel like a king whenever your sales reps are in town
As far as I can tell, selling B2B security products is mostly about making C-suiters feel like they're in a political thriller. They even build (and, at least when needed for these purposes, staff!) fake and useless "war room" sets to walk the guys through, inefficient dashboards that nobody doing the actual work uses because they suck but they look cool, stuff like that.
Oh I see you have not hung out with 'the sales guys'. That they did it on a motorcycle ride does not surprise me. There are some seriously shady dudes out there that will do anything to close the deal, "always be closing". If closing out a million dollar contract means the CEO's daughter wants to goto a water park? Magically there are the tickets aplenty. Bars, race tracks, sports games, horseback riding, Caribbean cruises, on and on.
Is it ethical? Not really. In fact many places have training specifically on vendor relations. How much, how little, etc etc. In a small startup environment? There are going to be basically no guidelines from the business to do anything either way. Larger companies tend to have guardrails. But many of the sales guys know how to work that system.
In technical roles we usually do not see this mess. Because we have a sense of follow the rules and logic. The sales guys are 'always closing no matter what it takes'.
on edit: Or Hell's Angels...
It is lead by Musk and I am sure he will use AI for that. Present it as "If Musk Fails, we will fail".
So many of them are obvious AI to anyone who has used LLMs. The emails are always like "Hi, I really like how <general fact about company mission> and how you used to <old job on LinkedIn>. We can 10x your business..."
Another fun one is they say they care so much about our business that they recorded a personalized video of them exploring our website. But the video is a person gesturing at their computer screen that only shows a Cloudflare bot blocking page because their AI video generator got blocked by our site as a bot.
It's so lame. It feels incredibly off-putting and dishonest that I am having my time wasted by a machine pretending to be a real person spending their time on me.
The problem is that this automation leads to the death of the entire sales channel. If 99% of "personal" emails I get are computer generated and the volume of emails keeps increasing because it's now so easy to send them, I'm going to stop reading any emails. I feel burned.
This is the problem with AI sales. It can automate the current average sales process. This in turn makes the average sales process really easy, so it gets saturated by everyone and then it no longer works for anyone.
If anything, you should do the opposite of whatever the AI sales people are currently doing. That's the way to make a mark.
I think this is one of these scenarios where feeling the pain is the only way they'll ever truly understand.
Heck, you can even A/B test it by randomly splitting your leads between humans and AI, then pick whichever has the higher ROI.
A better example (rather than replacing sales reps with AI, which is obviously difficult and nuanced) would be something like StitchFix deciding whether to fire their human stylists and moving everything to AI-recommendations only. Again, A/B test and see what users respond well to.
(Separately, I've always assumed StitchFix "stylists" weren't actual people, until I met someone with that job title... I was honestly quite surprised they still had a job)
And who cleans the following mess?
Normally we try to avoid damage - which frequently cannot be fully repaired.
And cleaning up mess is not an efficient use of resources.
Incentive is there to find non-damaging solutions first (Carlo Cipolla warns us about the amount of damage introduced in the system - the balance can be tricky). Cost-Risk-Benefit are due diligence.
...If a NN comes up with better solutions, it has won (the battle)...
The mess that is left behind is not damage. It's evidence. It's what would keep people in jobs. People losing their jobs is damage. People having work to do is not.
People having work to do which could have been avoided, resources wasted in repairs that were not needed - that's really what we do not want.
They should not be working then.
(I know they do work anyway even if they should not. What I am saying is, "the problems are at the source".)
If for «getting the result you want» you meant "creating damage in order to prove a point and discard some ways", I said that there must be a better way.
I'm not what you don't understand. People need experience before doing stuff with stuff that is completely brand new?! It's completely impossible. You're taking an approach which you think results in not being wrong. That is about you trying to be 100% right. And that attitude results in your being completely disregarded and those who you think are wrong doing stuff you don't want. You've just been a naysayer.
> If for «getting the result you want» you meant "creating damage in order to prove a point and discard some ways", I said that there must be a better way.
One you're talking about damage. If there is damage after 48 hours you're completely terrible are your job. It should just be a somewhat quick clean up. And there isn't. Anyone who has dealt with this sort of stuff and got the results they wanted knows sometimes you need to see the pain before you understand that something doesn't work. Most people need evidence to believe something. Your naysaying doesn't do it. You're not helping, you're just wasting time.
Not exactly. Once something come out you study it as much as it is necessary, you assess it, then you employ it smartly - that's how it should be done. Those "needing evidence" in your depiction seem akin to those who decide to date drug addicts, jump down buildings etc. because "they do not know yet, it could work out fine": the first problem that one has to pose is that those profiles should not manage - that should be the main background idea. Then, if the issue - at human resources level - happens, strategies must be implemented to contain damage. This implies, you do not talk about creating damage lightly. One bad experience for a counterpart makes you lose opportunities forever: the failed counterparts will go, "good to know, now next please".
> And that attitude results in your being completely disregarded and those who you think are wrong doing stuff you don't want
Actually, what you would do in real life, is proving your point to the interlocutor through showing the evidence ("Have you noticed <this and that>?"). I do not have to convince you here about any specific position such as "LLMs can't fly", it's not the relevant purpose here.
> naysayer
I have said, (1) perform a Cost-Risk-Benefit and see what is the best strategy; (2) do not just stand over the specific case but foresee where its roots bring, and work on them roots. That is positive.
The first way is diplomacy. Show and reason. If it fails, then artillery. But in that order.
(And before all of that, prevention.)
I don't think you've ever actually done any of this.
> The first way is diplomacy. Show and reason. If it fails, then artillery. But in that order.
Letting someone do a trial is diplomacy...
If you mean that the "normal way" I have drawn is not feasible, show that it is not. If you mean "they cannot be convinced", well... They must be either convinced or overridden, if you are the one that bears the damage.
My manager who has zero experience coding us now vibe coding all day.
Spoiler alert : it isn’t ending well
Production crashed, weeks debugging. Lots of wow I didn’t expect that
Why?
When some people hire, they have their subordinates sit in meetings all day, doing occasional tasks, and merely feeding their enlarged egos. If all you want are subordinates to feed your ego, AI is exceptionally good at that. Plenty of people love talking to chatbots.
The problem is the author never really explained what the roles were. Were they customer facing sales calls? Did the CEO really believe that customers will be happy to talk to a sales robot?
Thus, because I believe these roles aren't customer facing, I suspect that these roles are either feeding someones' ego by sitting in meetings all day, or otherwise non-customer-facing roles that handle aggregating information. This makes me wonder if a smaller group of people, who know how to use AI well, will outperform a larger group of people without AI.
> The "Replace or Justify" Ultimatum
It is hard to take this kind of stuff seriously. Actual businesses that produce value do not operate in this way.
I feel like one of the more important lessons I picked up along my journey is that ultimatums are a really bad idea. Instead of creating dialogue and exploring the entire gradient of in-between goldilocks solutions, you've narrowed an ~infinite spectra into 2 discrete, highly adversarial/tribal bins. This is not a good premise for a conversation around AI and how it applies to business. I don't know of a single business venture that couldn't extract some value from AI. Perverting this notion into an all or nothing narrative is so ridiculous to me.
There is no """AI""" ready to be entrusted...
And people should compete upwards, "try hard".
Fully automated company with no humans in the loop. That’s what’s coming.
If he said he "wants AI to do it all" then why shouldn't it include his job too?
Literally the point of the article was that they successfully did that
Snark aside, it sounds like your CEO doesn’t trust you. CharGPT can generate a report for you showing benefits. Or you can pay McKinsey a few hundred grand (if you’re lucky).
John has a special issue. He calls for support. Support is made of low paid personnel with no real knowledge of the product; during the call, they encouraged John by telling him that his issue is solved by - non-existent - features in the product, that the LLM aiding the "human interfaces to clients" matmulshitted.
John is rightfully upset for a number of reasons. (Happened a few weeks ago.)
Invented by longtime member Baq in a discussion with me after https://news.ycombinator.com/item?id=41602198
I am also partial to "stochfabulation" (which linguistically really is "inventing stories while guessing").
I’ve given up on resisting this idiocy and am just trying to stay out of the blast radius while I roast marshmallows on the ensuing fires.
Tell them you have an AI CEO and they have to meet the same standard. I can show you a demo :)
This sounds like an argument for turning it all over to AI to me. Both the buying and the selling.
Thing is, on a personal level yes, things like motorcycle rides are important. That is what life is made up of, what we live for. But a corporation should exist to provide goods and services to end users and a return to shareholders. This is an optimization problem and not a place for motorcycle rides. In theory (not arguing for or against capitalism) this optimization raises the standard of living for everyone.
Selling by personal connection and a motorcycle ride is really a form of cronyism or corruption. The end consumers and shareholders get shafted. This kind of nonsense is especially pronounced and explicit in less well off societies and lowers the general standard of living. So let AI do it. Oh, also give AI the manager who wanted to replace the salespeople with AI's job. That should probably be one of the first jobs to be replaced.
What a naive and utterly incorrect statement. A corporation is a peice of paper that separates ownership from liability. Attached to that peice of paper are some special tax rules, rules about structuring ownership and governance (particularly important when there are multiple owners) and so on.
Any notion of how they should be run is just a random person's opinion, unless you happen to have some percentage of ownership - then you get that percentage say in how things are run. Although I will say any sort of mechanistic opinion on corporate interactions is a broken one - fundamentally there are people doing things to fulfill an inter-corporate agreement, and some of those people will find ways to be shady or lazy or otherwise not live up to the agreement. The inter-personal relationships are very important, as they help the people running one corporation gauge how likely the other corporation will live up to it's end of the agreement.
Hardly. It's just "less than ideal."
> The end consumers and shareholders get shafted.
You're completely assuming that 100% a good deal CANNOT possibly be made on the back of a motorcycle? There's no reason to believe a suboptimal strategy can't arrive at the optimal solution.
> This kind of nonsense is especially pronounced and explicit in less well off societies and lowers the general standard of living.
This is the way business has been done in the first world for 100 years. You should calibrate your assumptions to the actual data.
`This is the way business has been done in the first world for 100 years.`. Yes, I am aware cronyism and low level corruption exist in the first world. In other places you probably just give the person an expensive gift or briefcase full of money and save time. But it's not optimal.
Point being, this story sounds like an argument in favor of AI to me.
You start your statement with a conditional. So, in your estimation, how often is it true?
> But it's not optimal.
And your assumption is that it would be cheap to optimize literally every decision we make regardless of it's total impact on the actual outcomes? It's almost never worth the actual costs for the given gains.
> this story sounds like an argument in favor of AI to me.
It's sounds like an argument for you to achieve more real world experience to me.
I'm not sure what your point is exactly.
I think my point is pretty clear but here it is again just in case. "If high dollar business deals are being made because of personal likes or feelings of the CEO towards a salesperson, AI may just maybe be more optimal".
10 years or so.
> I'm not sure what your point is exactly.
I disagree with you and am attempting to explain why.
> "If high dollar business deals are being made because of personal likes or feelings of the CEO towards a salesperson, AI may just maybe be more optimal".
If. What percentage do you think do? How big of a problem do you think this is /actually/?
Personal likes and feelings may play in but are you suggesting that's the only metric being used in these circumstances? Like if a company has a product that just won't work and will cause tons of problems you're saying a motorcycle ride could smooth that over?
What makes you think AI can optimize this problem that might not even actually exist?
It's amazing to me that people think AGI is going to come about while simultaneously doing exactly what you tell it to do. I mean I'm literally trying to picture the world you say you would prefer and I just can't. Worse even if I blur the definitions I actually can't see it being more "efficient."