When Daniel visits the US from Israel, he travels with an empty suitcase and fills it with Amazon orders—niche electronics and hard-to-find items he can't get at home. His experience highlights three pillars that make Amazon's marketplace feel like a different universe: selection, delivery, and returns. But each pillar casts a shadow. The packaging problem isn't carelessness—it's engineered conservatism. Fulfillment center software deliberately chooses oversized boxes and fills them with air pillows because a damaged item costs Amazon far more than extra cardboard. The returns problem is a direct consequence of the generous policy that makes the marketplace trustworthy. Bracketing—ordering multiple sizes and returning the ones that don't fit—drives a significant portion of the $24.7 billion Amazon spent on return-related costs in 2023. Many returned items are never restocked; they're written off or destroyed because reverse logistics costs exceed resale value. The labor problem compounds everything. Amazon's serious injury rate is 6.8 per hundred workers, nearly double the warehouse industry average, driven by real-time productivity tracking and a scheduling system that rotates workers through unpredictable shifts. Turnover exceeds 150% annually, meaning the company replaces its entire warehouse workforce every eight months. These three problems aren't independent—they feed each other. Fast picking quotas lead to poor packaging decisions. Generous returns increase warehouse volume, which increases pressure on workers. And the whole system is built on keeping prices low by squeezing labor costs. Solving any one piece requires addressing all three.
#4772: Amazon's Three Shadows: Packaging, Returns, and Warehouse Labor
How Amazon's selection, delivery, and returns create hidden costs in cardboard, waste, and worker safety.
Episode Details
- Episode ID
- MWP-4951
- Published
- Duration
- 26:57
- Audio
- Direct link
- Pipeline
- V5
- TTS Engine
-
chatterbox-regular - Script Writing Agent
- deepseek-v4-pro
AI-Generated Content: This podcast is created using AI personas. Please verify any important information independently.
Downloads
Transcript (TXT)
Plain text transcript file
Transcript (PDF)
Formatted PDF with styling
You Might Also Like
Never miss an episode
New episodes drop daily — subscribe on your favorite platform
New to the show? Start here#4772: Amazon's Three Shadows: Packaging, Returns, and Warehouse Labor
Daniel's in the US right now, visiting family. He does this every year or two, and every time, he flies with an empty suitcase and fills it with Amazon orders shipped to wherever he's staying. This trip, he was hunting for a niche WiFi 7 internal card for his laptop — something he couldn't find in Israel at all, but there it was on Amazon and Newegg. And he's not alone. There's a whole genre of Israeli tourist comedy built around stuffing suitcases with American goods. One skit has tourists packing the actual hotel lamps.
Wait, the hotel lamps?
The hotel lamps. The bit is that anything not bolted down goes in the suitcase. It's comedy, but it's built on a real price and selection gap that's genuinely enormous. So Daniel's been marvelling at the logistical machine — and he's also been feeling kind of bad about it. The cardboard. The returns that get tossed. The fulfillment center conditions. His question is whether you can keep the marketplace benefits — the selection, the delivery, the returns — while getting rid of the parts that are increasingly unsustainable. Is there a version of this that actually works?
And I think the way into this is to start with what makes it feel like a different universe to someone coming from a place like Israel. Because it's not one thing — it's three things, and they're connected in ways that make the sustainability question hard.
Selection, delivery, returns.
Those are the three. Selection first. Amazon has something like twelve million items in its own inventory, and when you add the marketplace sellers, you're looking at hundreds of millions of SKUs. For someone in Israel, where local e-commerce is fragmented and a lot of niche products simply aren't stocked, that's the whole game. Daniel's WiFi 7 card — that's an internal M.2 card, probably a specific chipset, specific antenna configuration — you're not finding that at the local computer store in Jerusalem. You might not find it anywhere in the country. But Amazon has seventeen listings.
And the delivery piece?
In Israel, packages go to a post office or a pickup point. You get a slip, you go stand in line. Amazon delivers to your door, often in two days or less. For Daniel, it's even simpler — he's shipping to a hotel, the front desk holds it, he picks it up when he checks in. The friction is nearly zero.
And returns are the third pillar. Print a label, drop it at a UPS store or a Whole Foods, money's back in your account before you leave the parking lot. In Israel, returns often mean arguing with a seller, paying return shipping, and waiting weeks.
That return policy is the trust engine. It's what makes someone willing to buy a WiFi card they've never touched, from a seller they've never heard of, knowing they can send it back if it doesn't work. And that brings us to the three shadows. Each of those pillars has a cost that's built into the structure, not just tacked on.
Let's go through them. The cardboard first.
So here's what most people get wrong about Amazon packaging. They think it's carelessness — someone in the fulfillment center grabbing whatever box is closest. It's the opposite. The fulfillment center software chooses the box size based on an algorithm that's deliberately conservative. It prioritizes damage avoidance over material efficiency. If there's any doubt about whether an item fits safely in a smaller box, the system bumps it up and fills the gap with air pillows.
So the excessive packaging is a feature, not a bug.
It's engineered in. The math is straightforward. A damaged item costs Amazon the product cost, the return shipping both ways, the processing labor, and potentially a lost customer. A few extra cubic inches of cardboard and some plastic air pillows cost a fraction of a cent. The algorithm is doing exactly what it was told to do.
And they have made some progress on this. I know they've reduced per-shipment packaging weight by forty-three percent since twenty fifteen, and they've eliminated something like two million tons of packaging material.
Right, and eleven percent of items now ship in their own container — no added Amazon box at all. That's the Ship in Own Container program. But eleven percent means eighty-nine percent of shipments still get the box-and-pillow treatment. And two million tons eliminated is real progress, but it tells you how massive the baseline was. The residual waste is still enormous.
What about the returns? Daniel mentioned the bracketing thing — people ordering three sizes of the same shirt and returning two.
This is the second shadow, and it's a direct consequence of the return policy that makes the marketplace trustworthy. The CBC Marketplace did an investigation on this — they found that thirty to forty percent of clothing returns are deliberate multi-size orders. People aren't gaming a loophole. The system is designed to make this behavior rational. Free returns, no questions asked — of course you order all three sizes.
And the cost of that?
Amazon reported twenty-four point seven billion dollars in return-related costs in twenty twenty-three. That's billion with a B. And here's the part that makes it a sustainability problem, not just a business problem: a huge portion of returned items never get restocked. They get written off, liquidated, or — in some cases — destroyed. The reverse logistics chain is expensive. Inspecting, repackaging, and relisting a returned item often costs more than the item is worth. So a shirt that someone ordered in three sizes and returned two of — one of those returns might end up in a landfill without ever being worn.
And then there's the third shadow. The fulfillment centers.
The injury numbers are the headline here. AP News and a Senate committee investigation found that Amazon's serious injury rate is six point eight per hundred workers, versus three point six for the warehouse industry average. Nearly double. And these aren't paper cuts — we're talking ergonomic injuries from repetitive motion, back injuries from lifting, injuries from being struck by objects. The productivity quotas are the mechanism. Workers are tracked in real time — pick rates, stow rates — and if you fall below the rate, you get flagged. The pressure to maintain speed leads to unsafe movements, bad lifting form, corner-cutting.
And turnover above a hundred and fifty percent annually, according to the Guardian's reporting from this past April.
Yeah. That number means Amazon is replacing its entire warehouse workforce every eight months. Think about what that implies for training, for institutional knowledge, for the human cost of people cycling through a job that chews them up and spits them out.
What I keep noticing is that these three problems aren't independent. They feed each other.
They absolutely do. The fast picking quotas lead to poor packaging decisions — a worker rushing to make rate isn't going to carefully select the optimal box. The generous return policy increases the total volume of items moving through the warehouse, which increases the pressure on workers. And the whole system is built on keeping prices low, which means squeezing labor costs. You can't pull on one thread without the others moving.
So let's talk about the packaging problem specifically, because I think the mechanism is worth understanding. You said the software chooses the box size.
The fulfillment center software assigns each order to a box size from a set of standard options. It's making a probabilistic call — what's the likelihood this item gets damaged in transit in this box size, given the weight, the fragility, the distance it's traveling, the number of times it'll be handled. If the risk is above some threshold, it bumps to the next size up. The threshold is set conservatively because the cost of a damage claim is so much higher than the cost of extra cardboard.
So when you open a box that's twice the size of the item inside, with enough air pillows to fill a bathtub, what you're looking at is a risk-management decision.
And the people who deal with the consequences of that decision are a specific role in the fulfillment center called problem solvers. These are workers whose entire job is fixing packaging errors — wrong box size, damaged items, items that shifted in transit because the box was too big. The existence of that role tells you everything. Amazon knows the software makes mistakes. It's cheaper to hire people to fix the mistakes than to make the software less conservative.
That's... wait. So they've got workers whose job is to clean up after an algorithm that's deliberately wasteful because being less wasteful would cost more in damage claims. And the waste — the cardboard, the plastic, the fuel to ship oversized boxes — that's all externalized. Amazon doesn't pay the full cost of that.
Right. The environmental cost of the excess packaging is borne by everyone. The damage claim cost would be borne by Amazon. So the algorithm does exactly what the incentives tell it to do.
What about the returns side? Is there a way to reduce the bracketing problem without killing the trust that makes the marketplace work?
There are a few approaches. One is sizing tools — Amazon has something called Fit Insights that uses AI to predict your size based on past purchases and returns. Third-party services like True Fit do something similar. If you can get the size right the first time, you don't need to order three. The problem is that these tools are only as good as their data, and clothing sizing is notoriously inconsistent across brands. A medium from one brand is a large from another.
And for non-clothing items, the bracketing equivalent is different. For Daniel's WiFi card, he's not ordering three — he's ordering one, but he's relying on the return policy as insurance against incompatibility. If returns weren't free, he might not buy it at all.
That's the structural tension. The return policy is what makes the long tail of niche products viable. A WiFi 7 internal card for a specific laptop model — that's an item that might sell a few hundred units a year. The seller can only offer it because the return policy gives buyers confidence. If you remove or restrict returns, you don't just reduce waste — you shrink the available selection. The two are linked.
And the return-less refund thing Amazon does for some low-cost items — where they just give you your money back and tell you to keep the product?
That's a pure economic calculation. If the reverse logistics cost exceeds the item's resale value, it's cheaper to refund and move on. It reduces some waste — the item doesn't get shipped back and processed — but it doesn't address the underlying issue, which is that the system is designed to generate returns as a normal part of the purchase flow. Returns aren't an exception to be handled. They're a feature of the customer experience.
Let's move to the labor side. The injury numbers are bad. The turnover is catastrophic. What's actually driving it?
The standard narrative is productivity quotas — and that's real. Workers are tracked to the second. Pick rates, stow rates, time off task. If you're not hitting the numbers, you're coached, then warned, then fired. The physical demands are intense — ten-hour shifts on concrete floors, repetitive lifting, walking miles a day.
But you've mentioned before that there's another factor that gets less attention.
The scheduling system. Amazon uses something called mega-cycle shifts in a lot of facilities — it's a scheduling model where your shift times can change from week to week. You might work days one week, nights the next, then a swing shift. For someone with childcare responsibilities or a second job or just a body that needs a consistent sleep schedule, that unpredictability is brutal. I've talked to people who said the physical work was hard but manageable — what made them quit was never knowing when they'd be working next month.
So even if you fixed the ergonomics and reduced the injury rate, you'd still have the turnover problem driven by scheduling chaos.
That's my read. And high turnover isn't just a worker welfare issue — it's a sustainability issue. Constantly training new workers is inefficient. New workers make more mistakes, which means more damaged items, more packaging errors, more returns. The human churn feeds back into the waste stream.
So we've got three problems that are structurally linked to the three benefits. Now the question Daniel actually asked: can you decouple them? Is there a version of this that keeps the selection, delivery, and returns while eliminating the waste, the return abuse, and the labor exploitation?
Let's take them one at a time and see what's possible. Packaging first, because it's the most straightforward. Amazon already has the Frustration-Free Packaging program and the Ship in Own Container initiative. The problem is incentives. Sellers have to redesign their packaging to qualify, and Amazon doesn't bear that cost — the seller does. So adoption is slow.
And the algorithmic box-sizing issue — you could change the risk threshold. Make the software less conservative. But then damage claims go up.
Right. The fix is technically simple and economically painful. You could also change the packaging materials — move to recycled and recyclable content, eliminate plastic air pillows entirely. Amazon's done some of this, but it's incremental. The deeper question is whether the distributed fulfillment model itself is the problem. Amazon's strategy is to have inventory positioned close to customers so they can deliver fast. That means items get shipped individually from local fulfillment centers in separate boxes. Compare that to a consolidated shipping model.
This is where the comparison to Alibaba's Cainiao network is interesting. We've talked about how Cainiao broke Israel's postal system — they flooded it with packages. But their model is different. They consolidate shipments. Multiple orders get bundled into one package at a hub in China, shipped together, and then broken apart for last-mile delivery. Per-item packaging waste is lower.
The tradeoff is speed. Consolidated shipping takes longer. Amazon's two-day delivery requires distributed fulfillment, and distributed fulfillment means more boxes, more trucks, more individual shipments. If you want the speed, you get the waste. If you want less waste, you accept slower delivery.
And for Daniel, who's visiting the US for a couple of weeks, the speed matters. He needs his orders to arrive before he flies home. But he's also the one looking at the pile of cardboard in his hotel room and feeling guilty.
The returns problem is even harder to decouple. The obvious fix is to charge for returns or restrict them — but that directly undermines the trust that makes the marketplace work for international shoppers and for niche products. If Daniel has to pay return shipping on a WiFi card that might not be compatible, he probably doesn't buy it.
What about the sizing tools you mentioned? If you can make bracketing unnecessary, you reduce returns without restricting the policy.
That's the most promising approach, but it only addresses clothing, and only partially. The broader problem is that free returns encourage a shopping behavior where the purchase decision happens after delivery, not before. You order first, decide later. Changing that behavior requires changing the policy that enables it.
And the labor side?
The debate here is automation versus regulation. Amazon has a billion-dollar Industrial Innovation Fund investing in warehouse robotics. The idea is that robots handle the repetitive physical tasks — lifting, moving, sorting — and humans do the more complex work that robots can't do. In theory, that reduces injuries. In practice, it changes the injury profile rather than eliminating it. Workers in highly automated facilities report different kinds of strain — standing in one place monitoring machines, repetitive fine motor tasks — and the quotas don't go away. If anything, the robots set the pace.
And regulation?
Several states have introduced Warehouse Worker Protection Acts that would require transparency around productivity quotas and prohibit quotas that interfere with safety or breaks. But regulation only works if it's enforced, and Amazon has shown it's willing to fight these laws. The California law passed, and Amazon's injury rate did come down in California facilities afterward — but it's still above the industry average.
What strikes me is that none of these fixes address the deeper structural issue, which is the long tail itself. The whole promise of Amazon is that it has everything. Millions of SKUs, including items that sell maybe a few dozen units a year. That long tail generates inherent waste — unsold inventory, packaging for rare items, returns of niche products that don't fit or don't work. A marketplace that aims to have everything might be inherently unsustainable.
That's the uncomfortable conclusion I keep coming back to. The three benefits Daniel identified — selection, delivery, returns — might not be separable from the three costs. The selection generates packaging waste because niche items don't benefit from efficient bulk packaging. The fast delivery generates waste because it requires distributed fulfillment. The generous returns generate waste because they enable bracketing and impulse purchasing. You can improve at the margins — better packaging, better sizing tools, better working conditions — but the core model has waste built into it.
A truly sustainable marketplace might require accepting less selection, slower delivery, or higher prices. Or all three.
For someone like Daniel, the value is specifically in accessing goods that don't exist in his home market. The WiFi 7 card. The thing you can't get anywhere else. A sustainable alternative for international shoppers might look like a consolidated buying service — a freight forwarder that aggregates purchases from multiple retailers into one efficient shipment. That's slower and less convenient, but it dramatically reduces per-item packaging and shipping waste.
Or a regulatory approach — product availability standards that reduce the need for cross-border shopping arbitrage in the first place. But that's a much bigger conversation.
Amazon has made some sustainability investments that are real but don't touch the core model. The electric delivery van rollout — a hundred thousand vehicles planned by twenty thirty. The Daylight trucking initiative. These reduce the carbon footprint of delivery, but they don't change the fundamental structure that generates the waste.
The packaging waste is a feature of the risk-management system. The return policy is a feature of the trust-building system. The labor model is a feature of the cost structure. Can you reform any of them without breaking the whole machine?
I think you can reform at the edges. You can reduce packaging waste by adjusting the algorithm's risk threshold and making sellers responsible for sustainable packaging. You can reduce return abuse with better sizing tools and maybe a modest restocking fee for high-return categories. You can improve warehouse conditions with better quotas and predictable scheduling. But the machine itself — the everything store with two-day delivery and free returns — that machine runs on waste. The waste isn't a byproduct. It's part of the fuel.
Hilbert.
Hilbert: Chester, Virginia. Twenty nineteen. I did a peak season at the fulfillment center there — November through January. Hired as a seasonal associate. Quit after a pallet of dog food fell on my foot because the pick rate meant stacking unstable loads and nobody had time to fix it.
Hilbert: Chester, Virginia. Twenty nineteen. I did a peak season at the fulfillment center there — November through January. Hired as a seasonal associate. Quit after a pallet of dog food fell on my foot because the pick rate meant stacking unstable loads and nobody had time to fix it.
That's... what was the pick rate?
Hilbert: Three hundred fifty units an hour. If you're below three hundred, you get a talking-to. Below two-fifty, you're gone. The dog food pallet was forty-pound bags, and the system wanted them stacked six high on a pallet that was rated for four. I knew it was wrong. I stacked it anyway because the scanner was counting.
Hilbert: Three hundred fifty units an hour. If you're below three hundred, you get a talking-to. Below two-fifty, you're gone. The dog food pallet was forty-pound bags, and the system wanted them stacked six high on a pallet that was rated for four. I knew it was wrong. I stacked it anyway because the scanner was counting.
The scanner counts everything.
Hilbert: Every pick, every stow, every second you're not moving. But here's the thing. The injury numbers are real — I saw people with wrist braces, back braces, the whole thing. But most people I worked with didn't quit because of the physical stuff. They quit because of the schedule. Mega-cycle shifts. One week you're on days, six a.m. to four-thirty p.m. Next week you're on nights, six p.m. to four-thirty a.m. The week after that, a swing shift that starts at two in the afternoon. You can't plan childcare. You can't hold a second job. You can't sleep.
Hilbert: Every pick, every stow, every second you're not moving. But here's the thing. The injury numbers are real — I saw people with wrist braces, back braces, the whole thing. But most people I worked with didn't quit because of the physical stuff. They quit because of the schedule. Mega-cycle shifts. One week you're on days, six a.m. to four-thirty p.m. Next week you're on nights, six p.m. to four-thirty a.m. The week after that, a swing shift that starts at two in the afternoon. You can't plan childcare. You can't hold a second job. You can't sleep.
That matches what I've heard from other sources. The turnover isn't just about physical demands.
Hilbert: The physical demands are real, but people who do warehouse work know it's physical. They expect that. What they don't expect is showing up Monday and finding out they're working nights starting Wednesday. The scheduling is what breaks people.
Hilbert: The physical demands are real, but people who do warehouse work know it's physical. They expect that. What they don't expect is showing up Monday and finding out they're working nights starting Wednesday. The scheduling is what breaks people.
You mentioned the problem solver role earlier — the workers who fix packaging errors. Did your facility have those?
Hilbert: Four of them, full-time. Their whole job was walking the floor with tape guns and replacement boxes, fixing shipments where the software chose the wrong size. Wrong size means too small — item doesn't fit, or it fits but there's no room for dunnage, or it fits but the box bulges and the tape won't hold. The software never picks a box that's too big. That's not the error it makes.
Hilbert: Four of them, full-time. Their whole job was walking the floor with tape guns and replacement boxes, fixing shipments where the software chose the wrong size. Wrong size means too small — item doesn't fit, or it fits but there's no room for dunnage, or it fits but the box bulges and the tape won't hold. The software never picks a box that's too big. That's not the error it makes.
Wait — the software only errs in one direction?
Hilbert: It errs toward small, because the algorithm's trying to minimize cubic volume. But it doesn't account for the fact that a human has to actually close the box. So the problem solvers fix the ones where the box is too tight, and meanwhile every other box gets the next size up with air pillows because the system learned not to trust its own tight-fit recommendations.
Hilbert: It errs toward small, because the algorithm's trying to minimize cubic volume. But it doesn't account for the fact that a human has to actually close the box. So the problem solvers fix the ones where the box is too tight, and meanwhile every other box gets the next size up with air pillows because the system learned not to trust its own tight-fit recommendations.
The waste comes from both ends. The tight boxes that need rework, and the loose boxes that are overcompensating for the tight-box problem.
Hilbert: The loose boxes are the system saying, I got burned on that SKU before, bump it up. And nobody ever goes back and tells the system it's being too cautious. The problem solvers fix the immediate error, but the algorithm doesn't learn from the fix. It just keeps making the same conservative call.
Hilbert: The loose boxes are the system saying, I got burned on that SKU before, bump it up. And nobody ever goes back and tells the system it's being too cautious. The problem solvers fix the immediate error, but the algorithm doesn't learn from the fix. It just keeps making the same conservative call.
That's a concrete example of the waste being engineered in. The feedback loop is broken.
Hilbert: The whole place is feedback loops that don't close. The pick rate causes injuries, injuries cause turnover, turnover causes training costs, training costs cause pressure to raise pick rates to maintain throughput. Nobody sits down and says, maybe if we lowered the pick rate by ten percent, we'd save more on turnover than we lose on throughput.
Hilbert: The whole place is feedback loops that don't close. The pick rate causes injuries, injuries cause turnover, turnover causes training costs, training costs cause pressure to raise pick rates to maintain throughput. Nobody sits down and says, maybe if we lowered the pick rate by ten percent, we'd save more on turnover than we lose on throughput.
Did anyone at your facility ever try to make that case?
Hilbert: The facility manager had a whiteboard with numbers on it. Throughput, cost per unit, injury rate. The injury rate number was always red. He'd point at it in the morning meeting and say, we need to get this down. Then he'd point at the throughput number and say, we need to get this up. Nobody asked how you do both.
Hilbert: The facility manager had a whiteboard with numbers on it. Throughput, cost per unit, injury rate. The injury rate number was always red. He'd point at it in the morning meeting and say, we need to get this down. Then he'd point at the throughput number and say, we need to get this up. Nobody asked how you do both.
That's the tension in one anecdote.
Hilbert: I still have the steel-toed boots. They're in a box somewhere. The dog food pallet bent the steel cap. Didn't break my foot, but I walked with a limp for three weeks. The safety guy filled out a report and nothing changed. The pick rate was still three-fifty the next day.
Hilbert: I still have the steel-toed boots. They're in a box somewhere. The dog food pallet bent the steel cap. Didn't break my foot, but I walked with a limp for three weeks. The safety guy filled out a report and nothing changed. The pick rate was still three-fifty the next day.
The thing I keep thinking about is what you said about the feedback loop not closing. The problem solvers fix the packaging error, but the algorithm doesn't learn. The safety report gets filed, but the pick rate doesn't change. The whole system is optimized for throughput, and everything else is treated as an externality — something to be managed after the fact rather than designed out.
That brings us back to Daniel's question. Can you keep the good parts without the bad parts? If the bad parts are externalities that the system is designed to ignore, then fixing them means changing the design. Not tweaking the algorithm — changing what the algorithm is optimizing for.
Hilbert: The algorithm optimizes for what it's told to optimize for. Tell it to optimize for something else, it'll do that instead.
Hilbert: The algorithm optimizes for what it's told to optimize for. Tell it to optimize for something else, it'll do that instead.
But the something else would be slower delivery, or higher prices, or less selection. That's the tradeoff.
Hilbert: Yeah. That's the tradeoff.
Hilbert: Yeah. That's the tradeoff.
The open question I'm left with is whether the tradeoff is actually as stark as it seems. If you reduced the pick rate by ten percent, injury rates drop and turnover drops and training costs drop. Some of those savings offset the throughput loss. If you reduced the packaging algorithm's conservatism by five percent, damage claims go up but packaging costs and waste go down. Some of those savings offset the damage claims. The system might be able to absorb some reform without breaking — not because the waste isn't structural, but because some of the waste is just bad optimization, not essential to the model.
As resource constraints tighten — packaging material costs, labor availability, regulatory pressure — the optimization problem shifts. What looked like an acceptable tradeoff in twenty nineteen might not look acceptable in twenty twenty-nine. The question isn't whether the marketplace changes. It's whether the change produces something better, or just a more expensive version of the same problems.
This has been My Weird Prompts. Thanks to our producer, Hilbert Flumingtop.
If you've got a weird prompt about a system that works perfectly except for the parts that don't, send it to the show — my weird prompts dot com. We'll take it apart.
We'll be back soon.
This episode was generated with AI assistance. Hosts Herman and Corn are AI personalities.