Hey!

Welcome to Flywheel, an exploration of micromobility, light electric vehicles (LEVs), and the technologies that enable them. This week’s edition features an interview with Ain McKendrick, the founder of Faction, about how tele-assisted supervised autonomy allows for truly driverless operations of LEVs that can deliver goods, and generate real revenue while doing so, today.

If you were forwarded this post and found it interesting, I would appreciate it if you subscribed to Flywheel!

Tele-Assisted LEVs with Ain McKendrick of Faction

Urban logistics is a hard business to make work. Delivery times are slow due to escalating congestion, costs are high, and the gig-worker model suffers from challenging unit economics in areas that aren’t incredibly dense. In response, California-based startup Faction is pioneering a practical blend of autonomy and human-assisted technology to deliver fully driverless, lightweight electric vehicles optimized specifically for urban delivery. Rather than pursuing complex, expensive autonomy stacks, Faction has engineered cost-effective solutions tailored to commercial micromobility, drastically reducing both hardware and operational costs. This week on Flywheel, I had the privilege of chatting with Ain McKendrick, founder of Faction, about how their supervised autonomy and right-sized vehicle selection enables truly driverless, revenue-generating delivery today. Please welcome Ain:

- - -

Puneeth: Thanks for taking the time to chat today Ain, it’s a pleasure having you here. To start us off, could you give a quick overview of your background and the origin story behind Faction?

Ain: Sure. My background is in consumer electronics; I spent many years there before moving into the vehicle space. I jokingly say I've always been the software guy at hardware companies. I've liked creating software-hardware-service type products because they tend to be more defensible.

Around eight years ago, I got into vehicles, initially automating systems for closed-course environments like ports, airports, and business parks. Although interesting, these are more industrial robotics-like—low-volume, highly customized—where economics can be challenging.

After that, I became VP of Engineering at Starsky Robotics, working on long-haul autonomous trucking. I still consider long-haul trucking one of the best autonomy use cases due to structured highway driving, a significant labor shortage, and rising demand. At Starsky, in 2019, we became the first to run fully driverless semi-trucks on US public roads, combining autonomy with teleoperation to handle ramps and toll booths. It reinforced my belief: you don’t need 100% autonomy to succeed commercially. You just need the right combination of technology and human assistance.

These experiences led directly to the founding of Faction. We focus on urban autonomous operations with remote assistance, evolving from direct teleoperation—steering wheels, pedals, high cognitive loads—to a third-generation approach, called TeleAssist. Vehicles handle driving basics autonomously (stay in lane, avoid collisions), while remote operators assist with edge cases: double-parked cars, construction zones, etc. The dirty secret in autonomy is that every autonomy company today relies on some level of remote assistance. Level 5 autonomy (fully generalized AI) has mostly been a red herring. At Faction, we deliberately balance onboard technology with human assistance in an attempt to prioritize economics and scalability. We're not a research project; we build practical, cost-effective, commercial vehicles.

Puneeth: What is the core Faction offering? Is it primarily a tele-assist kit, teleoperation services, full vehicles, or something else?

Ain: That's something that’s evolved quite a bit over the last few years. Initially, we thought we'd license our tech to OEMs. But OEMs aren't ready for this, and customers that need delivery/transport services don't want to manage fleets themselves. So we've landed on a robotics-as-a-service (RaaS) model. Customers prefer dealing with one company, so we handle everything: vehicle deployment, charging, maintenance, autonomy/tele-assist services. We charge monthly subscriptions and per-operation fees, primarily for logistics applications. Essentially, we're a B2B2C service provider rather than merely supplying tech stacks without a concrete business model.

Puneeth: You mentioned that Faction combines autonomy with teleoperation. What exactly does "TeleAssist" entail for operators? Are they directly controlling vehicles with remote steering wheels and pedals?

Ain: Instead of directly controlling vehicles, operators provide vectors or trajectories. This reduces cognitive load for operators, allows us to operate tele-op over cellular communication networks despite their latencies and deadzones, and enables one operator to manage more vehicles. Think of it as virtual air-traffic control. Vehicles autonomously handle most situations but alert operators when there are edge cases like double-parked cars. Operators briefly intervene to suggest course deviations, which the vehicle then autonomously executes. Each interaction creates training data, which then allows us to make more of the drive autonomous in the future.

Puneeth: Faction's chosen vehicle form factor is a three-wheeled pod car, which is unique compared to the rest of the AV industry. The big players like Waymo, Zoox, and (effectively) Nuro all use passenger cars, while other urban delivery companies like Starship or Serve use sidewalk robots. Why did you choose this form factor?

Ain: This was a very deliberate selection based on product requirements and choosing the right tool for the right job. Sidewalk robots operate slowly on chaotic sidewalks and are regulated at the city level (too much compliance work when scaling nationally), making them too complex for efficient and at-scale deliveries. We wanted to operate on roadways because roads are regulated at state levels and allow for speedier deliveries. That being said, larger passenger vehicles are excessively heavy and expensive for smaller payloads and fall under highly regulated automotive standards. This led to us choosing a three-wheel motorcycle-class pod cars: they’re fast enough to operate on roads, federally regulated as motorcycles (fewer standards than cars), operator-friendly, scalable, and cost-effective. The fact that our vehicles are three-wheeled actually means that neither us nor our operators even need a motorcycle license. I like to joke that bringing you a pizza in a 4,000 pound electric sedan is a little bit of overkill. Our vehicles get more right sized with the actual payloads that people are trying to deliver.

Puneeth: Are there other advantages to the motorcycle classification?

Ain: It mostly boils down to much simpler federal standards and more freedom in modifications. Motorcycle classification allows flexibility in how we fit our vehicles without heavy ADAS integration requirements common to passenger cars. It significantly simplifies our regulatory overhead.

Puneeth: What’s the process of converting a standard pod car into a Faction delivery vehicle?

Ain: Today and for the near-term future, we’re developing in-house drive-by-wire systems, sensors, and compute modules that we retrofit the pod cars with. In the long-term future, we’re hoping to license and transfer sensor-layout designs to OEM partners so that their vehicles roll off their production lines "Faction-ready." We envision developing a Faction-qualified supply chain eventually so that OEMs can directly source the appropriate sensors and compute modules on their own. This is ultimately what’s going to facilitate easier scalability. It’s a pretty similar model to how NVIDIA deploys their Orin automotive platform.

Puneeth: Could you talk a bit more about some of the early traction and customer work that Faction has done

Ain: We initially tested two markets: vehicle-on-demand and logistics. Vehicle-on-demand is a great application for solving urban mobility problems but the challenge is that you have to get to a certain fleet size before you can really have the right density in an area. Whereas logistics, if we can save a customer money and can get them a delivery done faster, it's a very clear path to revenue, and it's a very clear path to scale. So logistics is really where we’ve seen most of our early product-market fit and immediate, scalable economics. A big win for us was our pilot with Chick-fil-A at one of their Houston franchises, which is actually their largest franchise by order volume in the nation. This pilot helped us validate our operational model and prove that we can do fast deliveries, with high/reliable repetition, and with clear economic benefits versus gig workers. We found that we were able to do robotic delivery for under $10/delivery, which is genuinely cheaper than gig-delivery, has none of the hidden costs typical of gig-delivery platforms (raising the prices of menu items to funnel the delta back into delivery costs), and is faster and more reliable.

Puneeth: What were some of the other key takeaways from your Chick-fil-A pilot?

Ain: High utilization, seamless restaurant workflow integration, and enthusiastic customer acceptance. The pilot demonstrated that restaurants and customers accept autonomous delivery eagerly, and it demonstrated that we could effectively complete an entire order workflow from the moment someone places an order on the Chick-fil-A app to the moment it’s picked up by the customer from a Faction vehicle. We were able to complete this entire workflow in under 30 minutes, with the Faction delivery taking only 8 of those minutes (on average). The pilot validated our scalability, operational workflows, customer interaction models, and the overall viability of our model at scale.

Puneeth: How does Faction compare from a unit-economics or performance perspective to other delivery options like sidewalk robots, gig workers on traditional food-delivery platforms, or incumbent AV companies?

Ain: Sidewalk robots can't practically deliver beyond a mile or so due to speed, so they’re great for campuses but not for urban delivery. Some companies have tried using robots in bike lanes, but bike lanes are only marginally better than sidewalks (in terms of speed) and they’re still regulated at the city level. Gig workers, even with e-bikes, are expensive mainly due to the labor costs (account for ~64% of the costs). 

Incumbent passenger AV companies address this labor cost, but the vehicles themselves are far too expensive for small payload deliveries. You need a vehicle that can compete economically with gig workers. Our competition isn't a $400,000 self-driving EV; it's a driver with a 15-year-old car willing to work for about $2 per mile. This constraint meant our entire vehicle system had to be under $35,000, including vehicle and all technology. Eventually, we aim to get that cost below $20,000. To achieve this, we chose light electric vehicles that require smaller, less expensive battery packs, and technology choices that could realistically run on such vehicles.

Some competing on-road AV players built bespoke, custom, expensive mini-pod delivery vehicles, often overloaded with technology. While our chassis might start around $15,000, some custom competitor vehicles approached $100,000 due to excessive onboard technology, putting them far out of practical price positions. From the start, Faction decided not to reinvent the vehicle chassis. Instead, we partnered with OEMs, adding only enough technology to get the job done effectively and economically. By strictly defining engineering and product requirements, avoiding open-ended research projects, we ensure efficiency. Over-teching the vehicle doesn't just inflate costs—it creates power consumption issues. Traditional autonomy stacks on full-size EV SUVs used for robotaxis might draw up to 8,500 watts of power. Our system draws only around 500 watts. Even if offered free advanced autonomy systems by bigger companies, we physically couldn't fit or power them without rapidly draining our battery, possibly within 90 minutes without driving a mile.

Our autonomous vehicles ultimately eliminate both the capex and opex barriers most other delivery players face, and thus offer clear economic advantages. Our deliveries are faster, cheaper, and economically sustainable.

Puneeth: On the labor side of things, you still have some labor needs given that you have remote operators tele-assisting your vehicles. What's the rough ratio of number of vehicles per operator that you need to hit before you can reach your target unit economics?

Ain: This is actually one of my favorite parts about how we model Faction because we shamelessly steal from things in aviation like air traffic control. So let's use air traffic control as an example. When an airplane is departing an airport, they go through a series of controllers. First, they'll talk to a ground controller who positions them at the airport. Then they'll talk to a tower controller who gets them out of the airport. Then, once they’re en route, they're talking to what's known as a center controller. And the center controller is looking at huge regions (i.e. Northern California vs. Southern California). So the amount of interaction between a controller and an aircraft decreases as they get into the larger legs that they're operating on. We look at this in the same way for on-ground vehicles. The majority of our controllers are more like a tower controller. They're moving vehicles around a sector or in a predefined area. Once they get on a highway, the operators don't have as many frequent interactions with the vehicle. So the way to think about it is that in a dense urban area, it's maybe about one operator to every five vehicles. When you're on a highway, it's maybe one operator to every 10 vehicles or less even. For us, the one to five ratio is our sweet spot that gives us the right margins. And knowing that we’ll need fewer operators as we start doing longer means that we’ll eventually get better than one to five. That's why I like to make the joke that to be successful in this market, you only need to be about 85% autonomous, not 100%.

Puneeth: One of the things you’ve mentioned about the Chick-fil-A pilot is that you did delivery in a hub-and-spoke model, where Faction vehicles live at the Chick-fil-A store to eliminate that first leg of getting the vehicle to the pick-up point in the first place. Did you see high enough vehicle utilization rates given that the vehicles were parked at the store?

Ain: We learned from Chick-fil-A that we could get strong utilization as long as the store had high volumes of customer orders because customers repeatedly choose us over gig workers due to faster, cheaper service. This actually drove our utilization close to 100%. With high-volume restaurants, we initially expected to need 2-3 vehicles. But we eventually realized that some may need more (5+) to handle peak volume.

Puneeth: How have recent advances in context-aware AI impacted Faction’s strategy or tech stack? What do you think the broader implications of this are for the autonomy industry as a whole?

Ain: I'll start with Faction more directly first. With Faction, the most direct application we're seeing with context-aware AI is reducing the workload for TeleAssist operators. Human labor is always the cost center. Baseline autonomy capabilities—staying in a lane, not hitting things—have benefited significantly from advancements made by many others over the last decade. Humans are never responsible for braking or stopping vehicles; that's all handled onboard with visual cameras, thermal cameras, and radars providing triple redundancy. AI has typically struggled with judgment calls because it isn't truly intelligent—it's essentially advanced pattern-matching. Humans excel at assessing and solving unexpected problems.

By taking human learnings and applying them through the newer AI models, we provide suggestions to TeleAssist operators, increasing efficiency. Instead of spending thirty seconds interacting with a vehicle, operators might need to spend just two seconds by accepting an AI-suggested action.

It's similar to how people use ChatGPT to proofread emails. You wouldn't let ChatGPT manage your inbox unsupervised, but it’s effective at suggesting improvements. That's precisely how we're applying AI initially with TeleAssist operators—to reduce workload. Over time, if an operator consistently accepts a solution suggested by AI for a particular scenario, we might automate that action directly within the vehicle, similar to automatic braking or collision avoidance. Once a scenario is proven to be consistent through human validation, we can confidently integrate it onboard.

I believe we'll approach Level 5 autonomy quicker using this human-in-the-loop, data-validated model rather than simply adding more technology to the vehicle. This feedback loop of trust and verification with humans allows us to prove consistency and reliability clearly to regulators and the public.

Looking at the broader future of the industry, I'm excited by recent advancements, particularly Large Language Models (LLMs), providing greater context-awareness. For example, early autonomous vehicle models—say, ten years ago—used LiDAR point clouds fed into neural networks trained on millions of miles of data. While these systems began navigating effectively, they remained somewhat unreliable black boxes sensitive to sensor positioning. If sensors changed positions or vehicles, model performance degraded significantly.

What we really want is greater context-awareness separate from direct sensor reliance. If you ask ChatGPT basic road rules—like stopping on red, going on green—it knows the correct answer contextually, independent of sensor input. We similarly desire an intermediate step, extracting elements from sensors, then asking an AI model, "There's a red light or double-parked car—what action should we take?" This separates sensor data from decision-making logic and increases the scalability and robustness of the stack across many different vehicles or form factors.

Early autonomy models overly integrated with sensor feeds were brittle. A more generic, intermediate context-aware model trained separately from sensor specifics ensures adaptability to changes in sensors or vehicles. I see autonomy evolving toward a multilayered approach like this. Context-aware models will significantly reduce operator workload and eventually migrate onto vehicles directly.

Puneeth: These advancements seem to be a really big boost for startups like Faction as they compete with larger, better-funded players. One comment I’ve heard time and time again is that many of the incumbent AV companies are stuck with legacy tech stacks that were state-of-the-art circa ~2018 or so.

Ain: Precisely, and older neural networks and legacy autonomy models are rapidly becoming obsolete. This provides an advantage for startups like Faction, as we're not burdened by legacy tech stacks and can quickly integrate these modern, context-aware AI approaches.

Puneeth: What’s coming up for Faction in the next few years? What are you looking forward to and hoping to accomplish?

Ain: Our main focus is dramatically scaling fleets. We've validated core tech and economics; now we're refining fleet-management systems, integrations, customer experience at scale—hundreds, then thousands of vehicles. 

Puneeth: Awesome Ain, thank you so much for joining and sharing your insights.

Ain: My pleasure. It’s great to discuss real-world execution rather than hype. A core fallacy in autonomy over the past decade has been treating it as a general AI research project—trying to solve the "go anywhere, do anything" problem. Many failed to ask, "What exactly are we trying to accomplish, and what will customers pay us for?" True product-market fit demands precisely selecting technology, vehicles, and services to meet actual market needs. Faction's relentless focus on this practical approach results in solutions that are cost-effective and precisely targeted.

You can learn more about Faction here. This interview has been edited for clarity and length.

For more observations and resources on micromobility and LEVs, check out rideflywheel.com/resources.

That’s it for this edition. Thanks again for joining, see you next time!

- Puneeth Meruva