Hugo Troche

7 RECOMMENDATIONS FOR HARDWARE STARTUPS It’s hard to understate the difficulty of the v1 launch of a consumer hardware company. I see a ton of new HW startups coming out. Here are some counterintuitive recommendations… 1) Target a small market for v1, not a large market. You need a small number of people to love your product and this will earn you the right to build a v2. You have to understand your v1 will not make you a successful company. Your v1 needs to prove that you deserve to exist in the first place. In the case of WHOOP, we started with pro athletes who were willing to put up with our v1’s short battery life and flaky Bluetooth connections because they so desperately needed Sleep and Recovery data. A broad consumer base would have rejected it. 2) Be great at no more than 1-3 things. If you try to be great at more than that, you will build a hardware that is mediocre at a wide range of things. You won’t think it’s going to be mediocre, but it will be. 3) You want a small launch. Not because you want small sales but because you want to have a steady ramp. It’s much better to do 100k of sales steadily over 12 months than in 2 months with a huge drop off. Keep in mind you need to manage your manufacturing partner and that there’s nothing better or more efficient than steady and predictable. 4) CAC is your enemy. Referrals are gold. The magic of referrals is that they signal high NPS and they help you grow steadily. Again everything that helps you have a steady HW manufacturing plan is good. Building a referral program for v1 launch is a good use of your time. 5) Be excellent at Software. Turns out being good at both Hardware and Software is a hard thing and something that your BIG COMPANY competitor is actually not great at (excluding Apple here). For some reason, there’s such a magnifying glass on the Hardware development, that Software is then overlooked by consumer startups. Don’t make that mistake. You can be great at both. 6) You should spend an unreasonable amount of time and money on ONE aspect of your Hardware. Could be the design or a technical breakthrough. But something needs to be SPECIAL. In our case, we designed a modular battery pack that allowed Whoop to be charged without taking it off your body. This made our development timeline much longer, but it allowed us to achieve something that no other wearable has achieved since: 24/7 health data. 7) Finally, have an unrealistic belief in yourself and your team. You are not supposed to succeed and the odds are stacked against you. Building hardware takes confidence and you’re up against big players who will rip you off. We went so far as to leave a message for our competitors on our circuit board (pictured below). In the case of Amazon and others, the prophecy came true. If you’re building consumer hardware, I wish you nothing but success. It will be super hard, but super rewarding. Good luck 👊🏼 #consumer #hardware #wearable #startup #whoop

4,788

213 Comments

PSA: Stop using the original OpenAI CLIP weights. I see a lot of people using CLIP for multi-modal retrieval. What many people may not know is that there are 100’s of CLIP models now since OpenAI’s original release. These newer models offer significantly better performance across many benchmarks. For product search related tasks we recommend the following. + Fastest inference: ViT-B-32 [laion2b_s34b_b79k] + Best balanced: ViT-L-14 [laion2b_s32b_b82k] + Best-all round: xlm-roberta-large-ViT-H-14 [frozen_laion5b_s13b_b90k] You might notice these are not the ones that perform the best on the benchmarks. Why? Because the benchmark only test for very specific tasks which are often quite different from search and recommendations. These are ones we have found to be the best for product related search and retrieval tasks and generally very good first choices if you are unsure where to start (even for non-product related search). Here is a hugging face space with all the data: https://lnkd.in/gxu5UYkt Read more: https://lnkd.in/g5_D4ccs

293

22 Comments

Notion vs Glean reminds me a lot of Tesla vs Waymo... Just like I think Tesla is going to solve the 'software' problem of autonomous vehicles before Waywmo solves the 'hardware' problem... I think that Notion is going to solve 'enterprise knowledge search' before Glean solves the problem of document and knowledge creation... Why? A couple reasons: 1) Just like Tesla, Notion has a huge advantage when it comes to the amount of underlying data and documents to train on... They already have 100M+ users, and likely 10x that in terms of the number of documents, files, etc. in team's workspaces. 2) It's very hard to change people's behaviour re: where/how they create 'work'. It's why a lot of companies have come and gone in that category... As exciting as Glean's AI writing assistants might seem, most people are just fine (or even prefer) a good old fashioned editor/writer. I could be way off here, but this will be an intersting one to follow. Thoughts? Where am I getting it wrong? #ai #enterprise #knowledge #management #notion #glean

3

1 Comment

Who's got a Waymo experience to share? I just had my first truly bad one. First, I should say I love Waymo. It's not in many places yet, but it's poised to be everywhere soon. The app tells me I've done 26 rides covering 97 miles in SF and mostly it's been an exceptional experience. Apart from some quibbles like getting dropped off a block away up a big hill, it's a zen-like experience and a smooth better-than-human-driver experience on the way to wherever I'm going. But I think we're all bound to experience some hiccups as we journey to the future. In my case, I was taking a trip into the Financial District for an event and I had 6 minutes to spare given the ETA. Then all of the sudden it decides it wants to pull into the left lane, which is filled with parked cars. I sit at first amused at this decision. It pulls up right behind the parked car and waits. Then it goes through a dance of deciding it wants to get back in the original lane (everybody loves to snub a driverless car in traffic!) and pulling ever closer to the parked car ahead. Miraculously the driver of the parked car shows up, loads his trunk and departs with ease. The Waymo decides the next course of action is to pull directly behind the next parked car in the lane. My amusement fades. I call support via the touchscreen. My 6 minute buffer that had evaporated to 1 minute suddenly tells me I'll be 13 minutes late, a driver assistance team is dispatched and on their way. Yikes! Then magically (I think they manually took control), the Waymo behaves like a normal driver and inches back into the normal lane and I'm on my way, arriving 1 minute late to the meeting. So in the end, a bit of time stress for someone who really likes to be on time (I started a time company after all) and not really world-ending. In fact, despite this unfortunate experience, I'll be loyally riding again. And so I guess this ends as a parable about what Product Market Fit looks like. If you can mostly deliver superior experiences, there's room for rough edges. (pictured below, mom's first trip in a Waymo)

28

4 Comments

$2 H100s: How the GPU Bubble Burst Last year, H100s were $8/hr if you could get them. Today, there's 7 different resale markets selling them under $2. What happened? The market has flipped from shortage ($8/hr) to oversupplied ($2/hr), because of reserved compute resales, open model finetuning, and decline in new foundation model co’s. Rent instead. https://lnkd.in/gpQEidmC

2

1 Comment

Reimagining Healthcare Workforce Platfforms: Turning Capital Into Labor The future isn’t just about filling shifts. It’s about scaling labor intelligently, driving better outcomes for systems, staff, and patients alike. Andreessen Horowitz and Alastair (Alex) Rampell reinforced this Vision in a recent post and it applies to the next phase of healthcare workforce evolution. Healthcare workforce marketplaces plus AI are at the tipping point of transformation, evolving into systems of record that don’t just manage labor—they optimize it. Much like EHRs centralized patient care data, these marketplaces can centralize workforce operations, aligning talent supply with clinical demand. But there’s a bigger opportunity here: turning capital into labor. Here’s what I mean. Historically, industries (including healthcare) have focused on enterprise computing tasks—streamlining data entry, compliance, and back-office functions. While valuable, this approach often stops short of transforming the real-world, repetitive processes that make up the bulk of work in healthcare. Think beyond enterprise software. The next frontier lies in leveraging embodied AI and workforce marketplaces to automate and enhance labor-driven processes, transforming capital investments into productive labor outputs. For healthcare, this means turning staffing costs into scalable, dynamic systems that deliver value far beyond filling shifts: 1. Dynamic Labor Matching: Capital invested in marketplace platforms enables real-time matching of labor supply with patient demand, replacing outdated scheduling models with predictive, automated systems. 2. Workflow Automation: From credentialing to compliance and shift bidding, embodied AI can streamline traditionally manual tasks, freeing human resources for higher-value work. 3. Labor as a Strategic Asset: Workforce marketplaces shift the focus from static staffing budgets to proactive labor deployment, ensuring that every dollar invested in labor drives outcomes and efficiency. It is about the billions of repetitive, planned tasks that AI can improve. In healthcare, that means automating everything from staffing logistics to clinical task assignments, creating a horizontal system of engagement for labor (think Zendesk) that integrates seamlessly with vertical, workflow-specific platforms (think Toast for staffing operations). The result? Hundreds of billions in untapped market potential—and a healthcare workforce model that transforms how care is delivered. By turning capital into labor through smarter marketplaces and automation, we don’t just solve staffing challenges; we redefine what’s possible in healthcare operations and ShiftMed is leading the way with health systems moving beyond tired models. #capitalintolabor #systemsofrecord #healthworkforce #aiworkforce #healthsystems #hospitals #cno #hospitalceo #healthcareinnovation

11

2 Comments

Exactly a year ago, I found myself in the back seat of a Waymo driverless car in Downtown San Francisco. I had watched countless videos of it before, but nothing prepared me for the reality of sitting there, watching the car navigating by its own. It felt like one of those dreams where you're driving from the backseat 😂 Today, Waymo is providing over 100,000 paid robotaxi rides per week in the US. Their vehicles have clocked over 14.8 million rider-only miles. Even more impressive, the Waymo Driver is 3.5 times better at avoiding crashes that cause injuries and 3 times better at avoiding police-reported crashes compared to human drivers. https://lnkd.in/e_VKa_CB #TechInnovation #AI #DriverlessCars #Waymo

33

The cost of training top models is increasing 2-3X per year, and at this rate will reach $1 billion per model by 2027. The findings are from Epoch AI and follow their recent analysis of increases in compute for foundation models running at 4-5X per year. They've also provided some interesting data on the relative breakdown of cost components like staff, hardware, and energy. A few thoughts after reading this: - I would assume staff costs won't increase as much as hardware and energy costs, so the relative contribution of human labor to future AI models will decrease, in yet another sign of capital's increasing dominance over labor in the industry (just look at NVIDIA's ever-rising stock price). - If compute is increasing 4-5X per year but cost is increasing 2-3X, is it safe to assume that training runs are getting more efficient? - Some people, like Anthropic CEO Dario Amodei, have said we'll see $1 billion training runs in the next year. Does this mean the 2-3X pace is about to increase? That Epoch is wrong in its estimate? Or that Amodei is overstating the cost, possibly to put fear into competitors who can't keep pace? - Only a few companies are going to be able to train future models. We're likely going to see further consolidation. Perhaps we get to a future with only two large general purpose models, like the Windows/MacOS and Android/iOS duopolies we've seen in the past. Article here: https://lnkd.in/g5SejXen

7

Great post from Will Ahmed at WHOOP. We are in the fortunate position at Memfault to work with many innovative hardware companies - at all stages in their development lifecycle (including Whoop). We see first-hand how hard launching these products can be. It’s the reason Memfault exists. If I were to add a couple of additional recommendations to Will’s list they would be: 1. Agree on a limited number of very clear signals for product quality that you can measure consistently, that everyone on your team, top-to-bottom, understands, and that are closely linked to the core value proposition of your product. These measures should act as the early warning system if something isn’t as it should be - they should be the benchmark for go / no-go decisions. "Is device performance acceptable across these key measures?" Yes/No. And if the answer is yes - that should be enough to ship with confidence, even if you know other things aren’t perfect. 2. Users are extremely unpredictable and there is no substitute for real-world data. With all the skill in the world (and even a heavy dose of good luck) there is ZERO chance you can predict upfront exactly how your users will actually use your product. Of course, we are biased here. But seriously - how you think your product will be used, and how it’s actually used, can be dramatically different. And this in turn can have a significant impact on what is deemed acceptable for reliability and performance.

27

With Waymo announcing the company is now conducting 50,000 paid rides per week in 3 cities, we decided to add some economic context to the announcement. Based on our analysis, Waymo is generating anywhere between $500,000 to just over a $1,000,000 a week in revenue depending on the average fare. Add in 2 more cities/expansions and Waymo is generating north of $100 million a year. How long do you think it will take Waymo to achieve $100 million a year in revenue? Let us know in the comments below. #RoadToAutonomy #AutonomyEconomy #Waymo #RoboTaxi #RideShare #AutonomousVehicle #Autonomy #Google

56

We're trying a new thing: Company spotlights to help people better understand what various health IT vendors actually do 🔦 We’ll be cutting through the marketing speak to give short, to-the-point descriptions of a company’s product and/or service + some context on how they see their differentiators. Tag any companies in the comments you'd like to see us cover (and follow Elion if you want to stay in the loop!)

35

13 Comments

The demand for AI Engineers is exploding - but the talent pipeline for AI engineers? Bone dry. Why is this important? Because this is not just a trend - it's shaping the future of technology. What’s fueling the rise for this demand? AI Adoption Surge: → 63% of companies are racing to embed AI into their products and services. Severe Talent Shortage: → Globally, only 200 experts are capable of training new AI models.  New AI Hubs Emerging: → Cities like San Francisco, Seattle, and New York are leading the charge in AI innovation and recruitment. Salary Boom: → Senior AI engineers are in high demand - Top SaaS companies are offering upwards of $925,000 to secure top talent. So, what’s the bottom line? → AI jobs now make up 22% of all U.S. software development job postings → In some states, there are 10 job ads for every qualified AI professionals → 75% of enterprise software engineers will be using AI code assistants by 2028 The AI talent acquisition race is intensifying - Strategic hiring approaches will be the ones shaping the tech landscape of tomorrow. #AIEngineers #TechRecruitment #SaaSHiring #TalentAcquisition

59

5 Comments

#BigData is Dead, Long Live #BigData! The conventional wisdom in AI is that more compute equals better performance. But this is no longer true. We’re entering an era where #data, not #compute, is the real #bottleneck. To push #AI forward, we need smarter data strategies, not just more power. Thanks for the insights, Harry Stebbings and @Arvin Narayanan! #AI #BigData #Innovation #DataScience #MachineLearning #Tech #Lakehouse #Iceberg

7

1 Comment

Grammarly just acquired startup Coda (last valued at 1.2bn) with a goal of a possible IPO. This could be a big move in productivity, especially for product teams. Coda CEO Shishir Mehrotra was recently on Lenny's podcast (first comment below). https://lnkd.in/gmgfcm2P

1

1 Comment

10 TRILLION $ QUESTION: How much of a leap forward will ChatGPT5 be? If it’s only 20% or 30% or even anything less then 100% better, it will indicate that these models are topping out and we are shifting into a deployment phase with roughly the existing capabilities. This has massive implications for the foundational model builders and infra providers, as we are reaching diminishing marginal returns from new foundational models. The investment cycle in new models will slow down dramatically and it will be a race to the bottom on token prices. Most of the value will accrue to the application layer. We are still in the early days of deploying this technology and haven’t yet found the new ai first applications that create $100b+ companies, but we likely will soon. Maybe even $1tn or even $10tn. If we find out ChatGPT5 is just incremental, most of the energy and capital will likely shift toward building the application layer racing to win this prize. If ChatGPT5 is >100%, or even 10x better than 4, it will be clear that larger models are not facing diminishing marginal returns and all bets are off and any and every investment dollar will scramble in to larger and larger models. This will include the hyperscalers, financial institutions and likely even most state actors. We will reorganize society around building orders of magnitude larger models, requiring rethinking energy policy, capital, geopolitics, data, rare earth supply chains, advanced chip manufacturing etc etc etc. Waiting for OpenAI to reveal ChatGPT5’s leap is the only thing that matters in the world right now. Get out your 🍿 Which one are you betting on?

6

5 Comments

Mohit gives a good intro and overview of bufferbloat, codel, cobalt, and PIE: https://lnkd.in/gwQNjwFh

26

1 Comment

CEO Alexandr Wang made waves with his announcement a couple of weeks ago about adopting an #MEI approach at Scale AI. Whether or not you agree with him on this, what rings clear is an unflinching commitment to hiring the best talent possible and some of the strategies used to accomplish this. I use him as an example because even though I am a recruiting and hiring expert, I learn a lot from those out in the trenches building their companies and he has obviously done so with great success! One of the strategies mentioned by Alexandr is "casting a wide net" in recruiting. I have become a firm believer in this strategy and it is one that we employ on behalf of clients. Here are just some of the reasons to cast a wide net in hiring: ✅ Casting a wide net is more likely to attract candidates with diverse thinking styles, problem-solving approaches, and experiences. This type of cognitive and experiential diversity can lead to increased innovation and creativity and can help your organization be more resilient to future changes in the industry or economy. ✅ A wider search can uncover candidates with unique skill combinations or non-traditional backgrounds who might be overlooked in a narrow search. These "hidden gems" can bring fresh perspectives and innovative solutions to your organization. ✅ A larger candidate pool increases the chances of finding individuals who are not just qualified, but are an excellent fit for both the role and the company culture. This can lead to better performance and reduced turnover. ✅ Engaging with a wider range of candidates increases your company's visibility in the job market and naturally expands your company's network. This can also lead to valuable market insights, potential partnerships, or customer connections that extend beyond the hiring process itself. Even candidates who aren't hired can become brand ambassadors if they have a positive experience, potentially leading to future hires. These are many of the reasons we cast a wide net on behalf of clients. Mainly we want them to have the assurance that they are hiring the best candidate possible and not having to settle. Even with all the benefits, I recognize that casting a wide net makes recruiting more labor intensive and time consuming. It also requires a more systematic approach to candidate management and may require advanced skills in assessing candidate fit. Whether for an executive or mid-level hire, we can handle the tedious, time consuming aspects. In addition to our full service model, we offer alternative solutions for non-executive level recruiting and hiring. I'd be glad to hear your needs and discuss ways that we can be of support. I'm also always willing to offer advice or ideas that you can employ on your own. A rising tide lifts all boats. Message me, let's talk.

5