The easiest way for agents to use your robotics data 🤖
Sydney, Australia
Joined August 2025
- Tweets90
- Following25
- Followers1.7K
- Likes97
Alloy Robotics retweeted
had fun chatting with @IlirAliu_ on how we scale robots beyond demos
all the magic (and challenges) are in deployments!
99% robot reliability sounds impressive. Until you scale to 10,000 actions a day.
@_joe_harris_ , Founder & CEO of @AlloyRobotics, explains why reliability becomes one of the biggest challenges as robotics companies move from prototypes to real-world deployments.
His company is building the AI Robotics Engineer, using AI agents to investigate failures, analyze robot data, and help engineering teams operate larger fleets.
Founded in Australia, Alloy has raised A$16M+ from Square Peg, Blackbird, Airtree, and Skip Capital, and now works with companies including Andromeda, Advanced Navigation, and SwarmFarm Robotics.
Joe previously helped scale Eucalyptus, which was later acquired in a deal valued at over $1 billion.
In our latest @22Astronauts_ podcast, we discuss robotics infrastructure, raising venture capital, China's role in robotics, the future of software, and what it takes to build a company in Physical AI.
This episode is brought to you by @CoreWeave, providing the specialized cloud infrastructure powering physical AI and large-scale model training.
𝗧𝗜𝗠𝗘𝗦𝗧𝗔𝗠𝗣𝗦
00:00 Building the AI Robotics Engineer
03:25 Why customers buy solutions, not platforms
09:30 Scaling Eucalyptus and lessons from startups
17:30 Founding Alloy Robotics
22:45 Raising venture capital and startup storytelling
28:00 Building products customers actually want
36:30 China's robotics ecosystem and global supply chains
44:00 AI coding and the future of software
47:00 Build vs. buy in robotics
51:11 Why 99% robot reliability isn't enough
Full Episode:
🎥 YouTube: youtu.be/ufa9lJ01mF8
🎧 Spotify: open.spotify.com/episode/0k3…
Alloy Robotics retweeted
99% robot reliability sounds impressive. Until you scale to 10,000 actions a day.
@_joe_harris_ , Founder & CEO of @AlloyRobotics, explains why reliability becomes one of the biggest challenges as robotics companies move from prototypes to real-world deployments.
His company is building the AI Robotics Engineer, using AI agents to investigate failures, analyze robot data, and help engineering teams operate larger fleets.
Founded in Australia, Alloy has raised A$16M+ from Square Peg, Blackbird, Airtree, and Skip Capital, and now works with companies including Andromeda, Advanced Navigation, and SwarmFarm Robotics.
Joe previously helped scale Eucalyptus, which was later acquired in a deal valued at over $1 billion.
In our latest @22Astronauts_ podcast, we discuss robotics infrastructure, raising venture capital, China's role in robotics, the future of software, and what it takes to build a company in Physical AI.
This episode is brought to you by @CoreWeave, providing the specialized cloud infrastructure powering physical AI and large-scale model training.
𝗧𝗜𝗠𝗘𝗦𝗧𝗔𝗠𝗣𝗦
00:00 Building the AI Robotics Engineer
03:25 Why customers buy solutions, not platforms
09:30 Scaling Eucalyptus and lessons from startups
17:30 Founding Alloy Robotics
22:45 Raising venture capital and startup storytelling
28:00 Building products customers actually want
36:30 China's robotics ecosystem and global supply chains
44:00 AI coding and the future of software
47:00 Build vs. buy in robotics
51:11 Why 99% robot reliability isn't enough
Full Episode:
🎥 YouTube: youtu.be/ufa9lJ01mF8
🎧 Spotify: open.spotify.com/episode/0k3…
Alloy Robotics retweeted
> be andromeda
> build humanoid robots
> not for a demo video
> for actual aged care homes
> fleet grows 4-5x in six weeks
> waves and waves of bugs
> appoint a "goalkeeper" to stop every shot
> week-long shift too exhausting
> plug in alloy
> when abi stops mid-session
> alloy answers before any engineer does
> fleet-wide audio bug: 2 days -> 10 min
> finds a bug across robots nobody knew was there
> l1 & l2 support: automated
> scale more deployments, faster
(ep 2 of real world robots)
"robotics isn't ready yet"
we're making a series showcasing robots working in the real world 🦾
ep1: halter, valued at $2B & backed by peter thiel
Alloy Robotics retweeted
this changes who builds robots
you don't need a research lab to give your robot a brain anymore
for robotics teams without a big ML team, this is a game changer:
1. no more data plumbing to go from 0->1. you don't have to build ML infra, compute and data pipelines just to adapt a model to your robot
2. new skills in days. almond spent four days iterating with the API and now their robot deburrs the spacers used to build itself!
3. it works outside the lab. actor has a wheel loader filling dump trucks with nobody in the cab
4. you run the loop yourself. collect demos -> train -> test -> use the failures to pick the next data. faster loops -> faster learning -> better robots
it could be as simple as uploading your data, them training it, your robot running on it
its still v early. pi flagged network issues on some cloud-hosted deployments, which anyone running robots on spotty field 4G will know well
but if anyone can give their robot a brain now, the edge will increasing become deployment chops. exciting times!
Alloy Robotics retweeted
just as brains inspired the neural networks behind LLMs, human movement in judo, gymnastics and rock climbing can inspire how humanoids move
This paper visualised: "Multiple contact planning for minimizing damage of humanoid falls" (Ha & Liu, IROS 2015) turns falling into a planning problem.
- Model fall as a chain of inverted pendulums
- Solve for contact order with dynamic programming
- Minimise largest single impact, not the total
- Fall like a judoka!
Alloy Robotics retweeted
the 2026 robotics infrastructure stack:
- simulation and testing (NVIDIA Isaac Sim, Gazebo, MuJoCo, Antioch)
- connectivity and comms (Wi-Fi / 5G / LTE / Starlink, VPN backhaul)
- fleet management, monitoring and teleoperation (Formant, InOrbit, Open-RMF)
- robot middleware and messaging (ROS 2, DDS, Zenoh)
- navigation, planning and control (Nav2, MoveIt 2)
- project collaboration and sharing (GitHub, Tnkr)
- edge compute and onboard AI (NVIDIA Jetson, Qualcomm RB series)
- deployment and OTA updates (Ubuntu Core, balena, Mender)
- visualisation and replay (Foxglove, Rerun, RViz)
- CV/ML model training and labelling (Roboflow, Scale AI)
- robot learning and policy training (Hugging Face LeRobot, NVIDIA Isaac Lab)
- data upload from robot to cloud (AWS IoT Greengrass, Alloy)
- data logging, storage and management (mostly S3, Alloy)
- data pipelines and model delivery (custom ETL, CI/CD, registries)
- observability and root cause analysis (metrics, traces, Alloy)
- agent harness for robotics data (Codex, Claude Code, Alloy)
Alloy Robotics retweeted
> be swarmfarm
> world-class robots, built in australia
> raise A$46M
> 300+ robots deployed
> ticket queue up 10x
> plug in alloy
> day-long investigations -> 10 minutes
> 0 new support hires
> build australia made real
Alloy Robotics retweeted
"robotics isn't ready yet"
we're making a series showcasing robots working in the real world 🦾
ep1: halter, valued at $2B & backed by peter thiel
How do you move cows with software?
Draw a boundary on your phone. Halter's collars guide the cows using sound and vibration, with an electric pulse if those cues are ignored.
The company raised $220M at a $2B valuation to bring this to more farms.
First in our Frontier Robotics series, exploring how robots are being used in the real world.
Alloy Robotics retweeted
we barely marketed this event
but 300+ robotics people showed up to our massively oversubscribed party
there's a huge appetite to build hard things here in australia
Alloy Robotics retweeted
things every serious robotics team ends up rebuilding once the robots leave the lab:
- the sim → test → logs → replay → eval loop for policies
- observability across ros 2, CAN, fieldbuses, video and sensors that ops can actually use
- fleet management thats slightly more duct-taped than pure ssh + spreadsheets
- a data layer tying missions, bags, firmware versions, incidents, labels and models together
- audit-grade safety cases, traceability and operating procedures
- tools that let support and field teams diagnose, teleop and recover without opening a terminal
robotics won't have its akeoff moment until each of these have off-the-shelf solutions that just work.
Alloy Robotics retweeted
Engineers at OpenAI, Anthropic, Tesla and Waymo all backed the same Sydney startup 🤖
🔧 @alloyrobotics works out why a robot broke
📊 A robot throws off a gigabyte a minute. 1% is useful
⚡ Advanced Navigation went from a full day of analysis to 10 minutes
💰 $8M USD at an $80M valuation, led by Square Peg
Tesla and Waymo built this in house. Sydney sells it.
#anztech #robotics #australianstartups #ai
Alloy Robotics retweeted
We've raised $8M and now we're hiring across 7 different roles.
Our usage has grown 51% MoM every month this year so we're growing our world class team.
Open positions:
- Account Executive
- Chief of Staff
- FDE
- Full Stack Software Engineer
- Growth Generalist
- ML Engineer
- Pitch Your Own Role (Member of Robot Staff 🤖)
We obsess over our customers, pareto distribution everything, and iterate as fast as possible.
If this sounds like you, or someone you know, apply or reach out: app.dover.com/jobs/usealloy
We just raised $8M at a $80M valuation, to build agents for analysing robotics data.
The round was led by @SquarePegCap, with angels from @Tesla, @Waymo, @AnthropicAI, @OpenAI and more.
Alloy gives robotics teams and AI agents a shared memory over the data their machines produce.
We have processed data from more than 10,000+ robotic runs across almost 1,000+ robots, in several form factors across agriculture, maritime, drones, defence, humanoid, medical, logistics.
We understand the robot as a complete system across hardware, firmware and software, then bring answers into the tools where engineers already work.
The principle is simple:
- Every run should make the next one better.
- Every test should improve the next design decision.
- Every firmware update should carry the context of past fixes and regressions.
- Every support ticket should draw on what the fleet has already experienced.
That is the compounding loop we are building.
Raising capital is a means to an end. Helping teams build safer, better and more reliable machines is the goal.
This is our life’s work: accelerating robotic intelligence and, ultimately, robotic abundance.
If you’re building and deploying robots, my DMs are open.
If you want to help build this future, we’re hiring.
Alloy Robotics retweeted
We just raised $8M at a $80M valuation, to build agents for analysing robotics data.
The round was led by @SquarePegCap, with angels from @Tesla, @Waymo, @AnthropicAI, @OpenAI and more.
Alloy gives robotics teams and AI agents a shared memory over the data their machines produce.
We have processed data from more than 10,000+ robotic runs across almost 1,000+ robots, in several form factors across agriculture, maritime, drones, defence, humanoid, medical, logistics.
We understand the robot as a complete system across hardware, firmware and software, then bring answers into the tools where engineers already work.
The principle is simple:
- Every run should make the next one better.
- Every test should improve the next design decision.
- Every firmware update should carry the context of past fixes and regressions.
- Every support ticket should draw on what the fleet has already experienced.
That is the compounding loop we are building.
Raising capital is a means to an end. Helping teams build safer, better and more reliable machines is the goal.
This is our life’s work: accelerating robotic intelligence and, ultimately, robotic abundance.
If you’re building and deploying robots, my DMs are open.
If you want to help build this future, we’re hiring.
Alloy Robotics retweeted
ridiculous unfair advantage in robotics if you have made your robot+company data accessible to agents
Alloy MCP just launched.
Connect Claude Code, Codex, and other AI agents straight to your robot data.
Faster, verifiable, and far more token-efficient than feeding raw logs to the tools yourself.
Your fleet generates terabytes of recordings, and no agent can hold that in context.
Alloy MCP gives your agent a search engine over all of it. It finds the missions that matter and pulls back just the slice it needs, queried straight from your real data.
Watch the video to see it in action.
Alloy Robotics retweeted
tbh frontier robotics is just a data business.
except the data doesn't exist yet.
so you build it:
– robots + sensors are new senses. they turn dark corners of the world (infra, oceans, logistics, bodies) into data no one ever recorded
– models finally work. all that raw footage just sat there unused. models can now understand & reason across it at scale (alloy)
– every deployment emits more edge data. if you're clever, you find a way to get the customer to pay you and unlock negative collection cost.
– the hard part is the reward system. did the robot actually do the thing? get the verifier right and compound like crazy
– better models → faster, cheaper, harder deployments → more data. the loop tightens.
– and unlike the internet, none of this data is public. it's yours. and hopefully a moat?
over time the "robot logs" stop being telemetry and become the first written record of how the physical world actually works. and that trains the next models.
Alloy Robotics retweeted
the big problem in robotics:
teams are debugging with tools never built for them
its time for robots to get their own observability stack 🤖
Alloy Robotics retweeted
never been a more exciting time to be in robotics.
some insights from building @AlloyRobotics:
0:00:00 – engineer → atlassian → eucalyptus → alloy
0:09:30 – why now: cost curves + reusable rockets
0:10:45 – vertical farming post-mortems: unit economics, reliability, scale errors
0:13:40 – reliability in the physical world: from 99% to 99.999%
0:19:00 – operator-to-robot ratio as the core unit economic lever
0:23:30 – common data primitives (vision, time series, logs) + ROS-driven formats
0:26:15 – a robot produces more data in one minute than any LLM can hold in context
0:29:35 – edge AI tailwinds: jetson-class hardware, cheaper LiDAR + IMUs
0:30:20 – VLAs: perception → plan → act (and why smoothness matters)
0:42:15 – ideal customers: high throughput, real deployments, cloud telemetry
0:45:50 – robotics as a deflationary lever
Alloy Robotics retweeted
everything a robotics company needs to go from 0→1:
- one real customer problem with hard ROI, not a demo
- one hero workflow you own end-to-end (automation + human-in-the-loop fallback)
- simulation for fast iteration (NVIDIA Isaac, MuJoCo, Drake, Genesis)
- a single testbed robot you can abuse every day
- ROS 2 (+ Zenoh) + production-grade drivers and health checks
- a reproducible OS image and reliable OTA (Mender, Balena, Airbotics)
- logs and bags shipped to searchable storage (mostly S3 + custom scripts, Alloy)
- a tight feedback loop between operators, support, and engineering (Slack threads and prayer)
everything you need to go from 1→n:
- fleet management, task dispatch, and traffic control (Formant, InOrbit, Open-RMF, Cogniteam)
- secure connectivity with per-robot identity (Husarnet, Tailscale, Starlink backhaul, mostly custom)
- edge compute and onboard inference (Jetson, Hailo, Luxonis OAK, AMD Kria)
- observability wired into incident response and root-cause analysis (Nominal, Sift, Alloy)
- labelled data, evals, and a versioned ML pipeline into production (Roboflow for CV, W&B for tracking, custom ETL)
- safety cases, audit trails, and compliance your biggest customer will sign off on (mostly custom)
- multi-robot ops playbooks and a real on-call rotation (mostly tribal knowledge)
- a data layer that turns every mission into training signal (Alloy + internal glue)
there is variance across industries but these categories are broadly true for modern robotics teams
Alloy Robotics retweeted
Brand to web for @AlloyRobotics's launch.
The bolt mark is a nod to how robots were originally assembled. Every part of Alloy's presence feels precise, modular and engineered.