Owen Smith

I’ve been hearing / reading / seeing a lot about using synthetic users to validate ideas, test assumptions and do discovery. I’m genuinely curious… All I really see are declarations and debate about using them. I actually don’t care to engage in a debate about them because I don’t know enough. What I’m curious about is how synthetic users have the context they need to be useful. I talk to customers & users to get the context in their head, in their org, in their life etc… that I don’t have access to. That’s where the gold is. How do synthetic users convey that context without living it? Im genuinely curious. Would love some insight here from folks on the front lines.

11

2 Comments

Read my latest musings on the Product Breaks blog... I've put together my 2 min crash course on lagging and leading metrics. Featuring: ℹ️ Definitions 🚨Pitfalls 😁 Proactive tips to try Hope you enjoy! https://lnkd.in/eTt2EUhk

14

2 Comments

What uses more energy: asking an AI to review a complex technical spec, or boiling a couple of daily cuppas? 🫖🔋 In a nation that clears through roughly 15 million cups of tea a day, the kettle is our universal benchmark for power. But our digital workflows are rapidly catching up to our physical habits. Traditional search is beautifully optimized. Querying a classic like Cups of Tea brewed and consumed per day costs a minor 0.3 watt-hours per click. But when you hand a heavy document, pull request, or feature mapping over to Generative AI for a comprehensive review? A single complex prompt sequence can chew through anywhere from 1 to 30+ watt-hours depending on the model size and output length. Run a few of those reviews, and you've officially used more electricity than it takes to boil a fresh pot of tea. The productivity gains we get from embedding GenAI into our daily engineering and product cycles are incredible. It’s the ultimate 'oil' for smooth delivery. But multiplying those high-compute reviews by billions of global users means data center power demands are projected to double by 2030. Building and modernizing platforms for scale now means looking beyond standard performance metrics. Next time you're waiting for the kettle to boil, it’s worth a thought: Where does operational sustainability sit in your team's prioritization matrix? #ProductManagement #TechnicalProductManager #GenerativeAI #SustainabilityInTech #PlatformScale

3

This is post 2 of The PM's Guide to Claude. If you're just joining, the series is on my profile. The single highest leverage thing you can do before using Claude for anything else: Set up your working style file. Every time you open Claude without one, you're starting cold. No context, no sense of how you think, no idea who your stakeholders are. You spend the first few exchanges just getting it oriented. Then the conversation ends and you do it all over again. Write it once. Paste it at the start of any conversation, or load it into Claude Projects. Every interaction after that starts warm. One thing worth knowing upfront: this isn't a single file. It's a system of four. File 1: Working style file. About you. Changes rarely. File 2: Product context file. What you're building right now. File 3: Stakeholder map. Who you're navigating and how. File 4: Live context file. What's top of mind this week. Files 2, 3, and 4 I'll cover in the next set of posts on building your PM second brain in Claude Projects. Today, start with File 1. Takes 20 minutes. Do it before you open Claude for anything else today. One thing to keep in mind as you build it: everything in this file gets read by Claude on every conversation. That costs tokens. So keep it tight, every line should be something Claude would actually use. No filler, no context that sounds nice but changes nothing. Swipe through for what goes in it, with a real example from a monetisation PM, and a template you can copy directly.

10

1 Comment

Your CTO shipping code might feel reassuring. At scale, it’s often a sign the system isn’t working. We often celebrate a hands-on CTO. It feels safe. Execution is visible. Code is shipping. But there’s a quiet trade-off most scale-stage companies don’t notice immediately. Not because coding is bad. But because roadmap, strategy, and vision conversations slowly disappear. One early warning sign I’ve seen in scaling companies: When a CTO spends more time coding or reviewing PRs than leading discussions on where the system is going. The symptoms are subtle: • Senior leaders evaluated like individual contributors • Architecture evolving reactively instead of intentionally • Teams busy, but not aligned • Strategy quietly becoming “someone else’s problem” I’ve seen how this shapes behavior. When leaders are evaluated like ICs, they stop acting like leaders. They optimize for visibility, not outcomes. Marty Cagan captured this shift well: “Leaders are responsible for results, not for doing the work themselves.” Ben Horowitz framed it from a scaling lens: “The most important skill for an executive is the ability to get the right things done through other people.” The strongest CTOs I’ve worked with at scale weren’t the most visible coders. They were the ones creating clarity—around direction, trade-offs, and leadership beyond themselves. Coding can be a powerful tool—especially early on. But when IC-level contribution becomes the primary yardstick for leadership, the hidden costs show up later: plateaus, rewrites, attrition, missed windows. A question for CTOs, founders, and investors: At what point in your company’s growth did staying hands-on stop helping—and start limiting scale? #CTO #TechLeadership #ScalingStartups #EngineeringLeadership #StartupLeadership

50

9 Comments

Huge kudos to Mor (our very own Queen of AI, like Khaleesi, but with AI dragons), Natalie for her amazing design (and later Shany and Shira for your amazing design support), and of course Bogdan and his amazing R&D team behind this launch! And special thanks to Inbar for this amazing video and for being an incredible PMM :) AI Prospecting is where intent meets intelligence. Users can describe what they care about in plain language, and Similarweb’s AI searches and recommends the most relevant opportunities across the vast universe of data, filters, and signals only Similarweb can offer. Learn more here: https://lnkd.in/dJJixxPu It’s the first of several exciting new capabilities we’re unveiling this Q4 - and we’re just getting started.

35

If you know me, I don’t post very often on LinkedIn - and even more so those past few months as I took on the challenge to lead our Product team last November. I’ve always worked very closely to product, but actually being responsible for the team and what we deliver definitely brought its fair share of doubts, stress but also rewards. Like today :) This is the first big release we’re making since I’ve taken the role of SVP of Product and I couldn’t be more proud of the level of collaboration, engagement, creativity and passion I’ve seen across the company for this new release to come to life. I’m super excited to introduce Datamaran AI-powered Regulatory Monitoring! This wouldn’t have happened without the expert guidance of my colleague and self-proclaimed policy nerd Donato Calace <3 who has lived and breathed this for the past few months. 📌 3 things I’m most excited about: - With 39x increase in ESG regulations over the last 10 years and a proprietary regulatory database that we, honestly, haven’t touched much since I joined Datamaran 9 years ago. It was about time we gave this super valuable feature a lift! Including through… - our brand new AI Insights feature, built by our Innovation team led by Jérôme Basdevant. Love to see more AI embedded throughout our platform to supercharge insights and decision-making - More visuals! With over 19,800 regs tracked to explore and monitor, reg software doesn’t have to be boring :D Big shout out and thank you to the front-line team who worked on this Maclyn Afonso, Alexander Morgan, Mila Santos Orejón and José Carlos Garcés Calabria !! Interested in seeing it in action? Request a demo or learn more here: https://hubs.ly/Q044t28q0 #Datamaran #RegulatoryMonitoring #Sustainability #ESGRegulations #AI CC: Marjella Lecourt-Alma, Jean-Philippe Lecourt Ian van der Vlugt, Susanne Katus, Dennie S., Nick Geurds, Helen Skeen, Dónal Ó Mearáin Carly Pledge

73

15 Comments

Strong models are necessary. They’re not sufficient. There’s been a lot of excitement lately around new foundation model releases. That makes sense. Models are the base layer. But the part I don’t hear talked about enough is what surrounds them. In the human world, we’ve spent decades building buffers: mechanisms to slow decisions down, evaluate consequences, and assign ownership when things go wrong. That’s how complex systems stay stable at scale. With AI, those buffers are still thin or missing. Models are getting faster and more capable, but the surrounding systems checks, escalation paths, and clear accountability are not keeping pace. And that’s the real risk. Once AI is embedded into real decisions hiring, pricing, credit, healthcare, security the hardest problems stop being technical. They become questions of accountability, trust, and recovery after failure, not performance before a demo. That’s why I tend to respect teams that think beyond the model itself. Interfaces, incentives, rollout speed, and guardrails aren’t secondary concerns…They are the product. Models are the beginning but buffers, ownership, and trust are where the real work starts.

19

Been testing OpenClaw vs Hermes Agent. Interesting findings. I've been running multiple agents, with OpenClaw+Sonnet/Opus vs Hermes Agent+GPT5.4 (and more recently GPT 5.5). From a stability perspective, Hermes seems quite stable and it's a pleasure to update. Looks like they took note of everything that's wrong with OpenClaw and tried to address it. Out of the box, Hermes is more capable, but lacking in personality. It tries to convert everything to skills, whereas OpenClaw uses memory. It seems, memory wise, that it's more efficient than OC in its early days - but a mature OpenClaw installation typically has advanced memory configured. I used both with exactly the same skills - two different "heavy agents" that I've spent many weeks perfecting. Results: 1. Hermes is lacking on the connection front - it had bugs and dumb issues with Whatsapp (LID vs normal number) and Slack (it can't access information from the first post of a thread it was tagged in, but OpenClaw has slack issues too...). 2. Hermes agent is more stable with updates and credentials. 3. OpenClaw has more personality and seems to dive deeper into tasks and is more trustworthy. 4. Both write quite well, so for creation tasks, they are equally good 5. I wouldn't trust either one to write production code, but both can run Claude code + Codex. 6. Opus 4.6 still beats GPT 5.5 when combined with OpenClaw. 7. OpenClaw+Opus is faster and more in depth, but it's getting pretty close. So, for professional settings, where money is less critical and the tasks are complex, I would stick with OpenClaw+Opus. HOWEVER, I moved one of my agents to Hermes+GPT5.5 for research and writing purposes. I think with GPT advancement and with the rate of velocity - Hermes will catch up with OpenClaw on most major factors (that I care about) in the next month or so, if they continue on the same velocity. Right now, I'm running both on the same tasks (synced skills different machines) and both get triggered, so I can see how one does over the other.

4

Interesting night at Double Verify Impact yesterday. DV have long left being just a measurement platform, but seems to be positioning themselves as an all in Optimisation/Attribution platform, and even seem to be bringing AI to the private marketplace. The thing that still seemed missing to me though, as a proper Mobile Measurement partner. Over 70% of B2C sales are made on a smartphone in the MENA region, having that big of a slice of the pie being an "Integrated solution", or another partner, seems like an opportunity being left on the table.

16

Scaling doesn’t reveal how mature a product is. It usually reveals how much velocity was required to compensate for misalignment. There’s plenty of good advice about moving beyond feature delivery, being more intentional, clearer about outcomes, and more deliberate about how product work is framed. I see these as directionally right, but incomplete. In earlier product environments, progress often meant moving fast in the direction of whoever had the strongest pull. Work started quickly, and value was often obvious enough to justify after shipping. At larger scale, that stops working. Not because teams lose judgment, but because when many capable teams move in parallel, misplaced effort compounds fast, and the cost of being wrong becomes explicit. What changes most isn’t problem complexity, it’s that the value must be articulated before execution, not discovered through it. At a certain point, even expert intuition ceases to function as a coordination mechanism. Not because expertise disappears, but because intuition doesn’t scale across dozens of decision surfaces. That’s where maturity truly becomes evident. In more developed product environments, alignment isn’t seen as overhead: it’s regarded as essential product work, encompassing framing, shared language, explicit trade-offs, and clarity on what not to do. A somewhat unsettling conclusion I keep returning to: velocity looks like progress... until the product has enough blast radius for misalignment to matter. Photo by Ahmad Odeh on Unsplash.

16

I’m excited to share that we’ve launched a new feature designed to help customers save both time and money when shopping for items they’re interested in. Customers who have Alexa+ Early Access can now ask Alexa to automatically purchase items when prices drop below a target they’ve pre-set. Just search for what you want, then say something like, “Alexa, buy this for me when the price drops below $25.” Alexa+ will track the price and complete the purchase once it hits your desired threshold—so there’s no need to keep checking for deals or lower prices. This marks an important step forward in our mission to make shopping with Alexa even easier using the power of AI. I’m so proud of the team that brought this innovation to life and looking forward to what’s ahead!

152

14 Comments

Incredibly proud of what the Soter team has accomplished this year in driving innovation with SoterAI. Not just from an enhanced user experience to drive safer outcomes, but now with external validation of its accuracy and performance. SAFE-Bench was started to have an independent benchmark assessment performed on leading foundational models to see what AI performs best at safety & risk related tasks. Here are the results: ⭐ SoterAI Set a New Industry Standard: 🏆 SoterAI: 95.64% — Highest Overall  Next models: Gemini 2.5 Pro — 87.50% Claude 4.5 — 87.14% Grok 4 — 86.83% GPT-5.1 — 86.36% Copilot — 77.07% Where SoterAI Pulled Ahead 100% – Safety Intelligence 100% – Incident Investigation 95% – Hazard Identification 95% – Ergonomics 97% – RAMS 87.5% – Generative Materials 95% – Policy Review No domain below 87.5%. By comparison, Copilot went from 95% in Policy Review to 35% in Hazard ID. When it comes to performing multi-modal, complex reasoning tasks - SoterAI pulled ahead. What SAFE-Bench Reveals Foundation models do well on documents. They struggle with visual/applied safety tasks (Hazard ID avg: 66.25%). Specialized, domain-trained AI performs far more reliably. Why It Matters 1️⃣ Evaluate AI by task, not brand. 2️⃣ Domain-trained AI reduces miss rates and variability. 3️⃣ Visual safety is still AI’s hardest frontier, but SoterAI is incredibly performant. SAFE-Bench finally answers the question the industry has asked for years: Which AI actually performs best at real safety work? The data shows a decisive lead. SoterAI is setting the new standard for trustworthy, purpose-built safety intelligence. Full report: www.safe-bench.com

9

Most "AI features" today are horseless carriages—most Product teams just slap chatbots onto old workflows. But as Pete Koomen argues, true evolution requires a total rethink (https://lnkd.in/dPHVH4gd) My take: Intelligence should surface, not be requested. -I built this prototype to prove it. No chatbots. No prompts. -Context is the Prompt: Your GPS, images, and history are the instructions. -The World is the Input: Point the camera, and the app predicts the next valuable action. Seamless Flow: No "Hey Google," no typing—just subtle intelligence that knows what you need. We are moving toward a world that behaves like an operating system. #AI #UXDesign #Innovation #FutureOfTech

23

7 Comments

We've all seen the initial product launch checklists that initially look impressive. Few are actually useful. I see teams holding tight to giant launch plans - hoping complexity will protect them from risk, or blame. But what actually moves launches forward (and avoids all-hands-on-fire drama) isn’t a 70-point spreadsheet. It’s a one-page version that names the real constraints: Does the team trust each other to raise red flags early? Are the success criteria unmistakably clear, or have they been “inferred” one too many times? The hardest part of most launches isn’t what’s in the checklist. It’s the space between agenda items - where trust breaks down, or everyone’s slightly afraid to ask the obvious question. If you’re planning a launch right now: When was the last time you asked your team, “What are we most likely to miss here?” Did anyone actually answer? #productleadership #execution

3

I ran into this myself recently. I had to expense a conference ticket in CHF, regardless of the currency it was sold in. Buying the ticket was the easy part. The drag came after. I had to convert the price, then later justify which exchange rate I used to expense it at work. The rate on the day of purchase didn’t match the rate when I submitted the expense. Add credit card FX fees that often aren’t reimbursable, but you still want to use your card for travel and event insurance. Suddenly one click turns into unnecessary admin. That’s the kind of friction people don’t talk about, but it often shapes decisions. This is why pricing localisation matters. Paddle’s 2023 research showed localized pricing can increase conversion by around 30% versus showing a foreign base currency. Not because it’s cheaper, but because it removes uncertainty before and after the purchase. Introducing EUR and now GBP pricing at ZurichJS Conf is about reducing that drag. Clear pricing doesn’t just help you click buy, it helps you make your life simple and enjoy the experience. That’s part of our commitment to accessibility #ZurichJS #ConferenceDesign #DeveloperExperience #PricingStrategy #TechConferences #JavaScript #CommunityFirst

9

6 Comments

Watching designers unlock real MVPs with AI right now is genuinely breathtaking. For decades, great UX has been constrained by code dependencies, rigid frameworks, and the need to scale engineering teams before ideas could breathe. That bottleneck is disappearing. Designers and artists, people who deeply understand how humans want to interact with products, now move from concept to functional application on their own. . Not mockups in figma. Not prototypes. Real, working software. This is a fundamental shift. When the people closest to the user can also build, iteration accelerates, creativity compounds, and products feel more human by default. 💪 AI isn’t replacing creativity. It’s removing friction. And that changes everything about how great products get made.

27

2 Comments

"The irony of 'general purpose' tooling is that it pushes the hard work onto users. Opinionated design is harder for us builders, we have to make calls, accept tradeoffs, be wrong sometimes, and face the music (ie. people on X who have opinions). But that’s exactly why it produces better products, you’re deep in the weeds doing the hard work every day." https://lnkd.in/deMZKDuu

1

The most expensive bug I ever shipped was a billing script. Years ago, I was lead developer at a streaming startup and owned payments. For DACH subscriptions, the flow was awkward: signed agreements, subscription records, manual payment triggers, and a recurring script that extended subscriptions and charged active agreements. One day that script went feral. It started looping. Some customers were charged 30 or 40 times instead of once. There is no elegant engineering lesson when you are sitting there for days refunding people manually, answering angry messages, and trying to rebuild trust one customer at a time. It is just painful, humbling, and very real. But it did teach me something I still believe: The goal is not to build systems that never break. The goal is to build systems where failure is contained, visible, reversible, and boring. Idempotency. Limits. Dry-runs. Alerts. Kill switches. Audit logs. Clear ownership. The unglamorous stuff that sounds excessive until the day it saves you. This applies even more now with AI-assisted development. Faster building does not remove the need for operational judgment. It increases it. Every meaningful system will eventually do something stupid. Good engineering is deciding in advance how stupid it is allowed to be.

1