Are humanoid robots distracting us from the real unlock in robotics … hands? In this TechFirst episode, I dig into the hardest (and most valuable) problem in robotics: dexterous manipulation. Or, actually good robotic hands.

Guest Mike Obolonsky, Partner at Cortical Ventures, argues that about $50 trillion of global economic activity flows through “hands work,” yet manipulation startups have raised only a fraction of what locomotion and autonomy companies have.

We break down why robotic hands are so hard (actuators, tactile sensing, proprioception, control, data) and what gets unlocked when we finally crack them.

Watch (and subscribe to my YouTube channel):

The TechFirst podcast is also available on all the major audio podcasting channels …

Here’s a quick summary of the key points …

Robotic hands … the real unlock in robotics?

We love humanoids that walk and wave. (C-3PO, anyone?)

But after this TechFirst conversation with Cortical Ventures partner Mike Obolonsky, I’m more convinced that the major unlock in humanoid robotics isn’t legs … it’s robotic hands.

Manipulation is where the value is, where most jobs actually live, and where the hardest technical problems remain.

Comparatively, legs are easy.

$50T “hands work” vs. $15T “moving stuff”

• Mike’s core thesis: roughly $50 trillion of global economic activity flows through tasks that require hands and manipulation, compared to about $15 trillion for navigation/transportation.

• But the funding picture is inverted. Tens of billions have gone into autonomous driving and walking humanoids, while only ~$2–3B has gone to manipulation-focused “hands” companies.

Why the mismatch?

Autonomy has a clearer value prop, mature platforms (cars), and more structured environments (roads). Humanoids capture imaginations with locomotion demos and memories of science fiction movies.

Manipulation, by contrast, is messy, diverse, and brutally hard. And it’s easy to forget about, or ignore.

Why robotic hands are so hard

Human hands sit at the extreme edge of range (from surgeon-level precision to sledgehammer power) and feedback (touch, shear, vibration, pressure). To approach that, robotic hands need a stacked solution across hardware and software:

Actuation & mechanics

• Human-like tendons, joints, and a 3-DOF wrist in a compact forearm is non-trivial.

• The same hand must manage both fine, delicate motions and high-force tasks without breaking.

Sensing

• True tactile sensing (pressure, vibration, shear) is essential for grip, slip detection, and tool use.

• There’s no standard: piezoelectric, magnetic, capacitive, FSRs … each has tradeoffs, and the field hasn’t converged.

Proprioception & control

• Robots need robust internal sense of joint positions/forces to act reliably even with partial/occluded vision.

Perception limits in the wild

• Real homes and factories are occluded, reflective, wet/greasy, or deformable. Vision alone fails.

• Without tactile feedback, force estimation is wrong and reliability craters (cue broken dishes).

AI & data

• Teleoperation is often used to collect training data; fully autonomous dexterity is still early.

• Integrating new sensor modalities into policies that generalize across tasks is a massive data and modeling challenge.

Grasping ≠ manipulation

There’s a big difference between grasping and manipulation …

• Grasping is securing the object; manipulation is using it (e.g., cutting tomatoes, turning a screwdriver).

• Using tools extends the hand to the tool tip; humans “feel” through tools via vibration and force transfer—robots largely don’t yet.

• Many impressive humanoid demos (dishwasher loading, folding) highlight just how early we are on autonomous dexterity; some are still teleop under the hood.

Reliability is everything

Making robotic hands that work for a few weeks is one thing. Making them work for years is totally different.

• In warehouses and production lines, a key KPI is interventions per hour.

• Suction works until it doesn’t (weight, porosity, geometry). Boxes vary; bags deform; bimanual coordination is often required.

• A practical system balances autonomy with handoff to humans and assigns tasks based on capability and confidence.

Pragmatism vs. purity

Two development philosophies are colliding (and complementing each other):

• General-purpose hands: moonshot hardware + AI that can do almost anything a human can.

• Task-specific end-effectors: grippers/suction/tools that solve today’s high-ROI jobs.

• A realistic bridge: modular, swappable end-effectors so a robot can “change hands” for different tasks, and upgrade as tech improves.

Why locomotion led … and why manipulation must catch up

• Research, competitions (e.g., early DARPA focus), and clear benchmarks helped locomotion mature faster.

• But most useful home and factory tasks need hands. As Mike notes, assuming “general-purpose humanoids” also means general-purpose manipulation has been premature.

What happens if we solve robotic hands (even to the 70–90% level)?

In short: a lot.

• Productivity explosion across manufacturing, logistics, and services.

• New industries: from advanced assembly to delicate biotech handling to in-home assistance.

• Off-planet construction becomes realistic when hands can build, maintain, and adapt without humans inside the suit.

My take

We won’t match human hands soon.

But “good enough” manipulation for high-value tasks can fund the journey to great. Expect the near future to be hybrid: locomotion plus task-optimized end-effectors, teleop where needed, and steady expansion of tactile sensing + AI that gradually generalizes.

As investors and builders wake up to the $50T “hands work,” we’ll see the stack — actuators, sensors, control, data — start to compound.