Which Jobs Do Robots Start With? - Ali Zamani

# The Uncomfortable Truth: Robots Are Already Working

Let's dispel a myth first: automation is not something that will happen in the future. It is something happening right now. We just don't notice it because it's gradual and quiet.

Some real numbers:

750K+

Active robots in Amazon warehouses

40K

Workers replaced at Foxconn

100%

Chain stores with self-checkout

+10%

Annual industrial robot growth

What does this mean? It means when you use a self-checkout at a store, when your package was picked by a warehouse robot, when your food delivery navigates the street without a driver — you are directly experiencing job displacement. This is no longer theory.

Honest note: The goal of this article is neither to scare you nor to comfort you. The goal is to understand exactly which jobs are going, why they're going, and when. Because only by understanding the mechanism can you prepare.

# Why AI Alone Isn't the Culprit

When the media says "AI is taking jobs," there's a big problem: AI alone cannot do any physical work. A language model cannot lift a box. A neural network cannot mop a store floor. AI is only the "brain" — but a brain without eyes, hands, and legs is useless.

A functional robot is built from at least six layers of technology. Each must reach sufficient maturity before the robot can do a job:

L6

Actuators & Hardware — hands & legs

motors, grippers, mechanical arms

L5

Connectivity

5G, WiFi, low-power networks

L4

Edge / Cloud Computing

GPUs, TPUs, data centers

L3

AI / ML — decision-making

language models, reinforcement learning

L2

Computer Vision — perception

object detection, depth estimation, segmentation

L1

Sensors & IoT

cameras, LiDAR, IMU, tactile

This stack matters because it reveals a key insight: a job becomes automatable only when all six layers have matured enough. If even one layer is weak, the whole system fails.

For example, in 2015, AI could recognize images better than ever (layers 3 and 2), but LiDAR sensors were still expensive and bulky (layer 1), and robotic arms weren't precise enough for delicate tasks (layer 6). The result? Robots still couldn't fold laundry. Today, with cheaper LiDAR and improved soft arms, that wall is crumbling.

ⓘ

Key insight: Progress in any single technology isn't enough. When multiple technologies simultaneously cross a threshold, a job suddenly — and often unexpectedly — becomes automatable. This phenomenon is called technology convergence.

# Moravec's Paradox: Why Folding Laundry Is Harder Than Chess

In 1988, Hans Moravec, an AI researcher, articulated a strange paradox. He wrote:

❝

"High-level reasoning requires very little computation, but low-level sensorimotor skills require enormous computational resources." — Hans Moravec

In plain terms: for a computer, playing chess is easy, but folding laundry is hard. Recognizing faces is easy, but walking on uneven stairs is hard. Translating languages is easy, but opening an unfamiliar door is hard.

Why? Because abstract thinking (logic, math, chess) is a recent achievement in human evolution — only a few thousand years old. But sensorimotor skills (walking, grasping objects, balance) are the product of hundreds of millions of years of evolution. The human brain is optimized for movement, not mathematics.

Easy for Robots

Things that are hard for humans

Chess — robot wins since 1997
Mathematics — always wins
Data analysis — wins since 2015
Translation — nearly wins since 2020

Hard for Robots

Things that are easy for humans

Folding laundry — still in development
Walking on stairs — still unstable
Opening unfamiliar doors — still challenging
Repairing broken equipment — still complex

This paradox determines which jobs go first. Jobs that are mostly abstract thinking (accounting, data analysis, translation) get automated sooner. Jobs that are mostly sensorimotor skills in unpredictable environments (plumbing, nursing, cooking) go later.

But "later" doesn't mean "never." Every year that passes, the wall of Moravec's Paradox gets lower.

# The Real Map: Which Jobs Go First?

Now that we understand why some tasks are harder for robots, we can build a realistic map. Two main factors determine how exposed a job is to automation:

1. Task Repetitiveness: Is the same work repeated every day? Or is each situation new?
2. Environment Structure: Is the work environment fixed and predictable? Or different and uncontrollable each time?

First Casualties

Repetitive + Structured environment

Warehousing, assembly lines, cashiers, package sorting

Next Wave

Repetitive + Semi-structured environment

Agriculture, industrial cleaning, fixed-route driving

Automating Now

Non-repetitive + Structured environment

Accounting, translation, content writing, call centers

Safe For Now

Non-repetitive + Unstructured environment

Plumbing, electrical work, nursing, therapy

This map isn't static. Every year, as technologies improve, the boundaries shift. Jobs that were in the "safe for now" section slowly drift toward more dangerous ones.

Honest warning: "Safe for now" doesn't mean "safe forever." Even jobs in the green section are changing — not by being eliminated entirely, but by having parts of their tasks automated, reducing the number of people needed.

# Jobs That Are Being Destroyed Right Now

This is no longer a prediction. This is a list of jobs that are right now being automated:

Warehousing

Amazon Kiva — 750K+ robots
Amazon Proteus — autonomous carts
Amazon Sparrow — sorting arm

Cashiers

Self-checkout — 90% of stores
Amazon Go — no checkout
Smart item-detection cameras

Agriculture

Apple & strawberry picking robots
Precision spraying drones
Laser weeding robots

Transportation

Starship — food delivery in 50+ cities
Nuro — driverless delivery vehicles
Self-driving trucks on highways

Notice a pattern: in each of these cases, automation isn't the result of one technology. For example, an apple-picking robot needs all of these:

apple_picker.py

class ApplePicker:
    def pick(self):
        # Layer 1: stereo camera + LiDAR → apple position
        apple_pos = self.detect_apple()
 
        # Layer 2: vision model → ripe or not?
        if not self.is_ripe(apple_pos):
            continue
 
        # Layer 3: arm trajectory planning
        path = self.plan_trajectory(apple_pos)
 
        # Layer 6: soft arm + suction gripper → pick without damage
        self.arm.execute(path)
        self.gripper.suction_grab(apple_pos)

If even one of these layers didn't exist — for example, if suction grippers that can pick apples without bruising them hadn't been invented — the entire project would fail. This means automation is always the result of convergence, not a single technology.

# Jobs We Thought Were Safe — But Aren't

Moravec's Paradox said cognitive tasks would be automated first. This prediction was correct, but most people didn't take it seriously. Because it was assumed "cognitive tasks" only meant math and programming. That's not the case.

Cognitive tasks also include:

Accounting

AI reads receipts and calculates taxes

Translation

Language models surpassed human translation

Content Writing

AI writes articles, ads, product descriptions

Customer Support

Chatbots handle 80% of tickets

Legal Research

AI searches cases faster than lawyers

Medical Diagnosis

AI is more accurate than doctors on imaging

Graphic Design

Image generation, logos, initial concepts

Coding

Code generation, debugging, code review

This list is painful. These were the jobs we called "the smartest human work" ten years ago. Jobs that parents said "learn this, its future is secure." And now AI is doing a large part of them.

But there's a subtle point: automation doesn't necessarily mean complete elimination of a job. Sometimes it means one person can do the work of five. The result? The need for workers in that field decreases, even if the job doesn't disappear entirely.

Productivity ↑↑↑ ÷ Labor needed ↓↓ = Job count ↓

This equation is simple but ruthless. When a tool makes a worker 5x more productive, the company doesn't need 5x workers — it needs the same output. So instead of 5 people, 1 is enough. 4 jobs disappear.

# The Convergence Equation: Why Progress Is Sudden

One of the most important characteristics of automation is that it's not linear. Years may pass with nothing happening, and then suddenly a job gets automated in a matter of months. Why?

Because automation is a function of multiple variables that all must cross a threshold:

Automation(t) = min( AI(t), Vision(t), Hardware(t), Connectivity(t), Cost(t) )

The min function here is critical. It means automation isn't limited by the strongest technology — it's limited by the weakest one. If five technologies are excellent but one (say, cost) is still above the threshold, automation doesn't happen.

convergence_model.py

def is_automatable(job, year):
    # each technology has a maturity score 0-100
    ai       = maturity("ai", year)
    vision   = maturity("vision", year)
    hardware = maturity("hardware", year)
    cost     = affordability(year)
 
    # bottleneck: the weakest link in the chain
    bottleneck = min(ai, vision, hardware, cost)
 
    return bottleneck >= 80  # automation threshold

This model explains an important phenomenon: the sudden automation explosion. When all technologies gradually improve, bottleneck may stay below 80 for years. But when the weakest technology also crosses the threshold, suddenly is_automatable flips from False to True — and this can happen within a single year.

Real example: food delivery robots. In 2015, all technologies existed, but LiDAR was expensive ($100K) and edge image processing was slow. By 2020, LiDAR dropped below $500 and edge processing got fast. Suddenly, food delivery robots went from theory to reality. This wasn't an explosion — it was convergence.

ⓘ

Tipping point: The greatest danger to jobs isn't when one technology improves. The real danger is when the last bottleneck technology also crosses the threshold. That moment is sudden change.

# Jobs That Are Actually Safe — And Why

Now let's look at the green section. Which jobs are genuinely resistant to automation? And why?

They share three characteristics:

safe_jobs.py

def is_safe(job):
    # 1. Unpredictable environment
    unpredictable_env = job.environment == "chaotic"
 
    # 2. Fine motor skills in tight spaces
    fine_motor = job.requires_dexterity == True
 
    # 3. Emotional intelligence and human connection
    emotional_iq = job.requires_empathy == True
 
    return unpredictable_env and (fine_motor or emotional_iq)

Real examples:

Plumber / Electrician

Every home is different. Every broken pipe creates a new situation. Completely unpredictable environment.

Nurse / Caregiver

Human care requires empathy, touch, emotional understanding. A robot cannot provide comfort.

Creative Chef

Tasting, adjusting flavor, creativity. A robot can make a burger, but not innovative cuisine.

Therapist / Psychologist

Deep human connection, understanding emotions, building trust. The core of the work is being human.

But even these jobs aren't completely safe. Robots are already doing parts of their tasks: monitoring robots in hospitals, smart leak detection systems, AI assistants for therapists. This means the number of people needed decreases, even if the job doesn't disappear.

The "Last Kilometer" Problem: In automation, the first 90% is easy, the last 10% is unbelievably hard. A robot can do 90% of a nurse's job — monitoring vitals, dispensing medication, logging data. But that last 10% — calming an anxious patient, understanding their pain, making ethical decisions — will take years to automate.

# The 2027–2035 Horizon: What's Coming

If you think what we've seen is a lot, wait. Several technologies are maturing that could completely break down the wall of Moravec's Paradox:

Humanoid Robots

Tesla Optimus — target price $20K
Figure 02 — partnership with BMW & OpenAI
Boston Dynamics Atlas — electric version

Soft Robotics

Soft grippers — fruit, eggs, delicate objects
Artificial skin — tactile sensing for care

Swarm Robotics

Hundreds of small robots for farming
Inspecting infrastructure, pipes, bridges

Teleoperation

Remote surgery with zero latency
Elderly care by remote operators

The most important shift is humanoid robots. Why? Because the physical world is designed for humans. Stairs are built for human legs. Doors are designed for human hands. Tools are shaped for human fingers. A humanoid robot can work in the same space humans work, without any changes to the environment.

This is a tipping point. Until now, robots needed special environments (assembly lines, warehouses with QR codes on the floor). But a humanoid robot can enter any environment where humans work. This means the range of automatable jobs expands dramatically.

10

2010

Industrial robots
Assembly lines

20

2020

Warehouse & checkout
Early agriculture

26

2026

Autonomous delivery
Cognitive work

30+

2030+

Humanoid robots
General physical labor

But let me be realistic: the humanoid robot is still in its early stages. Walking on uneven ground, opening unfamiliar doors, working with delicate tools — these are still challenging. Predicting when they'll be "good enough" is difficult. But the direction of travel is clear.

# Honest Summary

Let's be honest. This article doesn't say "robots will take everything, doom is imminent" nor "don't worry, everything will be fine." The truth is more complex and more important than either:

1. Automation is real and happening now. This isn't the future.
2. But AI alone isn't the culprit. It's the combination of six technologies that makes robots possible.
3. Which job goes is predictable: repetitive work in structured environments goes first.
4. Cognitive jobs are also at risk — perhaps even more than physical ones.
5. Truly safe jobs are those with unpredictable environments + manual dexterity + emotional intelligence.
6. Progress is sudden because when the last bottleneck technology crosses the threshold, everything changes.

The biggest mistake is thinking about this as "AI versus humans." The truth is we're witnessing the biggest transformation of the labor market since the Industrial Revolution — perhaps even bigger. And like every major transformation, there will be winners and losers.

Preparation comes down to one thing: understanding the mechanism. Knowing why and how a job gets automated lets you reposition yourself before the wave hits.

takeaway.txt

              Robots don't take jobs — technology convergence takes jobs.

              And this convergence is like a slow flood:

              years of nothing, and then suddenly everything is taken.