I remember some years ago, in a warehouse far, far away… while on a site visit with an early-stage startup, watching three forklift drivers approach a junction at the same time.
No painted lines, stop signs, or written rules to see in the place (I’m not sure they’d heard of health and safety in this establishment). Yet, without hesitation, two of them paused. The third drove straight through.
“Steve always goes first,” someone told me. He’d been there the longest, so was the most senior.
That moment stayed with me. Because while others seem set on humanoid robots as a foregone conclusion, I think they underestimate the problem.
On paper, it seems simple. Build a robot that looks like us, walks like us, works like us. Put it in any setting: a warehouse, a factory, a care home. Let it take on repetitive tasks, maybe with a little tailoring. Preferably in the hazardous industrial environments we humans really shouldn’t be in. The demo videos make it look close to reality. What could go wrong, given the robot is like us?
But real environments aren’t demo videos. They’re messy. Someone leaves a box slightly out of place. A tool doesn’t go back on the right hook. A new employee doesn’t yet know that Steve goes first, followed by Steve T-800. Real environments are unpredictable, and delivering reliable dexterity, long battery life, and costs below a human worker is no mean feat.
Humans absorb all of this without thinking. We build a mental map of the space and update it constantly. We communicate with others. Find out who is senior, who is new, who looks uncertain. None of that is written down, yet we navigate it effortlessly.
For a robot, each of those adjustments is additional nuance that quickly becomes mountainous to codify and keep up to date.
Don’t get me wrong, we’ve made substantial progress. Simultaneous localisation and mapping (SLAM) gave us robots that can map environments and understand where they are within them. Perception using cameras or other sensors can distinguish between walls and glass, and between static objects and moving ones. Vision-language models allow robots to interpret environments in natural language.
But understanding a dynamic, hybrid human environment is not just a technical problem. It’s a social one.
You can avoid some of that complexity by structuring the environment, as Amazon and Ocado have done in highly automated warehouses. But the real opportunity, in my view, is building robots that can genuinely operate in hybrid human-machine environments and be truly assistive. It’s OK if they need to be tailor-made for a particular sector instead of being general-purpose for every environment, after all, most don’t expect a skilled manufacturing plant worker to also be a coding genius. However, what we don’t want are machines that need to be rescued by a human every time something unexpected happens.
Maintenance must be low and supervision minimal.
I’m pro-robotics, but I’m cautious about the humanoid hype. Hardware costs are falling considerably and the unit economics, while still speculative in pilot deployments, are beginning to look plausible. I want companies in the space to succeed, but there’s a tendency in technology to overfit. If you have a hammer, everything looks like a nail. Right now, the hammer is the humanoid robot. Especially considering the news about Tesla’s Optimus and NVIDIA positioning humanoids as the next AI compute wave.
We keep asking where we can use one, instead of asking what the task is and what the simplest, most robust way to do it might be. Could it be a wheeled machine or a quadruped with better mission-planning models?
The question isn’t simply whether we need humanoid robots. It’s whether they’re the right answer, right now. Much as I would love the help with all my house chores, not just the vacuuming!
Read on for more news from our portfolio and partners across Intelligence, Human, and Planet this month.
Intelligence
Refute raises £5m. We led the seed round to help Refute scale its counter-disinformation platform, expanding detection across new threat vectors and growing the team. Read the announcement.
Keeping multi-cloud data private. enclaive raised €4.1m to help teams run sensitive workloads across multiple clouds without exposing what’s inside. We backed the round through the Amadeus APEX Technology Fund. Read the announcement.
Quantum, without the hype. Carmen Palacios-Berraquero (Nu Quantum) and Steve Brierley (Riverlane) join Bloomberg to break down what “useful” quantum needs around the hardware—networking, error correction and security—with a clear takeaway: if your data must stay safe for 10 years, you’re probably already late. Watch the interview.
Quantum comms on today’s fibre. Photonic and TELUS have shown quantum teleportation running over existing network infrastructure, a concrete step toward quantum secure connectivity. Photonic also strengthened its board as it pushes toward commercial scale. Teleportation news. Board news.
A new lab for quantum networking. Nu Quantum opened a trapped ion networking lab in Cambridge to speed up the hands-on work behind distributed quantum systems. Read the announcement.
Making error correction less guessy. Riverlane shared a practical “error budgeting” approach (now in Deltakit) to help teams see what’s actually limiting QEC performance and where to focus next. Read the update.
Voice AI that behaves in the wild. PolyAI is rolling out with Fogo de Chão to handle every guest call, while a new Gordon Ramsay campaign takes aim at the phone experiences customers still hate. Customer news. Ramsay ad.
Security keeping up with AI scale. Sitehop published two sharp reads on why “good enough” encryption assumptions are breaking in regulated networks, and why crypto agility is becoming a near term operational issue. Read the posts.
AI for engineers, not instead of them. Secondmind’s CEO spoke with Entrepreneur UK on using AI to sharpen engineering judgement under real constraints, not replace it. Read the interview.Nu Quantum opened a trapped ion networking lab in Cambridge
Human
Protein to structure, faster. Nuclera and leadXpro are teaming up to speed structure-based drug work on notoriously difficult membrane protein targets by linking protein access with downstream discovery workflows. Read the announcement.
Rewriting life’s blueprint. In our In Focus interview, Constructive Bio CEO Ola Wlodek explains how writing whole genomes can turn cells into programmable biofactories for complex proteins and new materials—with far less waste than traditional chemistry. Watch the interview.
More reliable therapy manufacturing. A UKRI case study highlights how Innovate UK support helped Ori Biotech advance real-time monitoring within its IRO platform, targeting more consistent and efficient production of advanced therapies. Read the story.
Planet
Trusted trade gets a bigger table. Altana joined the Global Alliance for Trade Facilitation’s Global Business Partner network, focused on more transparent, data-driven trade systems. Read the announcement.
Recycling nylon 6,6 at commercial scale. INVISTA and Epoch Biodesign signed an MoU to advance post-consumer recycled nylon 6,6, combining Epoch’s enzyme work with INVISTA’s polymer and manufacturing footprint. Read the release.
Plastic free packaging on the show floor. At Packaging Innovations 2026, Huhtamaki and Xampla showcased folded cartonboard formats using Xampla’s Morro coating — built to keep fibre recycling viable while ditching plastic linings. Read the coverage.
A patent-led signal. Paragraf was named one of the UK’s ten most innovative startups by LexisNexis, a reminder that hard defensibility still shows up on paper as well as in product. Read the note.
Orbital logistics, fewer handoffs. Voyager and ATMOS Space Cargo announced a strategic partnership aimed at making it easier for customers to get payloads up and back without stitching together a dozen vendors. Read the release.
News from the team
Patient capital makes deep tech work. Writing for Tech Funding News, Dame Anne Glover argues that turning world-class science into global companies takes years, not quarters—and the UK needs growth-stage funding that matches that reality. Read the article.
Deep tech, in plain English. Edward Norton, Communications and Marketing Manager, shared a new guide on what deep tech is (and isn’t) — and what “proof” really looks like when you’re building something ambitious. Read the blog.
How to raise on proof. James Baker, Principal in our Early Stage team, published a practical guide for founders on raising pre-seed funding for deep tech. Read the guide.
Strengthening the UK’s translation pipeline. Partner Pierre Socha has been appointed to the Royal Society’s Science, Industry and Translation Committee. Read the press release.
If you enjoyed reading this issue of Capitalise, why not subscribe below to receive it directly to your inbox.