DiscoverTank Talks By Ripple Ventures
Tank Talks By Ripple Ventures
Claim Ownership

Tank Talks By Ripple Ventures

Author: Ripple Ventures

Subscribed: 12Played: 561
Share

Description

Join your host, Matt Cohen, Founder & Managing Partner at Ripple Ventures for weekly conversations with leaders in the startup ecosystem discussing the truth about investing, building and running startups.

tanktalks.substack.com
359 Episodes
Reverse
In this episode of Tank Talks, host Matt Cohen sits down with David Caputo, co-founder of Harmoni and a former RBC leveraged finance banker. After a decade in banking, David bought and operated aerospace and defense factories across the U.S. He saw firsthand how well-equipped plants still lose time, scrap expensive parts, and leak tribal knowledge, and that experience led him to build Harmoni. Harmoni is a factory orchestration system that pairs proprietary shop-floor devices with custom software to coordinate people and machines.They cover David’s path from electrical engineering to Wall Street to the shop floor, and the stories that convinced him to build rather than buy: a wrong-handed wing rib scrapped, a year of lost production because the only current program lived on a retired machinist’s USB stick, and a machine “cutting air” for a factory tour. David explains the gap in the ISA-95 stack where human operators leave the data pyramid, why he chose to build Harmoni’s MVP without his own factories as a testbed, and why owning the hardware was worth the pain. He also discusses Harmoni’s $10M Series A led by Bessemer, HAL (the context-aware manufacturing AI), and why modernizing the 300,000 existing U.S. manufacturers matters as much as greenfield autonomous factories.Whether you’re building in physical AI, investing in industrial technology, or curious about how AI is transforming the real economy, this conversation offers a ground-level look at the challenges and opportunities shaping the future of manufacturing.From Electrical Engineering to Investment Banking (02:24)* Growing up in Toronto and graduating into the post-dot-com tech slump* Sneaking into business school info sessions and landing in RBC’s Generalist Program* Helping build RBC’s New York leveraged finance practice from about 5 people to nearly 70* The VBA macro that earned him the nickname “Computer” by automating a multi-hour structured products comps updateBuying and Building Aerospace and Defense Factories (06:17)* Leaving banking in 2013 to start RVE Partners (Rift Valley Equity) and do lower middle market buyouts* The thesis: bring large-cap systems, data, and tooling to companies with under $10M of EBITDA* Building a platform that included a forge shop in Texas, a machine shop in Tulsa, and a composites shop in Massachusetts* The “oh crap” moment after closing the fund: raising money is just getting out of the starting gateWhat Factory Floors Actually Look Like (09:27)* Why value creation is much harder than it looks from the 100,000-foot view of banking* Landing gear plants running on scratch pads with no electronic devices on the floor* Why operators resist new tools: not job fear, but the burden of stacking 6 to 8 systems to enter the same information* Why more software can mean less efficiencyThe Stories Behind Harmoni (13:53)* Seventy people lining up at one terminal to punch in and then lining up again at another, and why it felt like The Flintstones* The left-hand instruction used on a right-hand wing rib, and why work instruction software doesn’t solve that problem* “Jimmy’s USB stick”: a year of re-engineering a part after the only current program left with a retired machinist* A machine “cutting air” during a factory tour, with no chips coming off the belt* Why there was no single final straw, just a long string of preventable problemsMeeting Co-Founder Adam Ellis (19:16)* Why David never intended to start a company and first wanted to buy software, then build an internal tool* Finding Adam through manager referrals and a blog post, then a cold email* One call that revealed the problem was far bigger than David’s own group of companiesWhere Harmoni Sits in the ISA-95 Stack (21:19)* A plain-English walkthrough of the five layers, from ERP down to the physical machine* Why the stack works for Toyota and Coca-Cola plants but not for most U.S. manufacturing* The U.S. landscape: roughly 300,000 manufacturers, about 90% with fewer than 100 employees, running job shops that change over multiple times a day* The core insight: no knob exists on a human operator’s head, so data leaves the pyramid and re-enters by hand, which David sees as the root cause of most inefficiency and scrap* Why the many-to-many problem (hundreds of ERPs, hundreds of machine types) has kept anyone from building a universal toolBuilding the MVP the Hard Way (26:13)* Harmoni’s three pillars: labor automation, process control, and observability* Why David chose not to use his own factories as a testbed, because he wanted real signals of efficacy* The home-heating analogy: a great furnace is useless if you leave gaps in the windows, so factory orchestration has to close all the gaps* Early customers who said the product was too early, and why that forced rapid iteration* The over-configurable dashboard that customers found too complicatedBuilding for Today While Reaching Toward Tomorrow (31:40)* Why outsiders can build breakthrough products in some industries, but manufacturing is a different animal* Wet-ink signatures, FAA rules, and OEM-mandated software that can’t integrate with anything* How Adam’s decades of manufacturing consulting and David’s operating experience shaped the productWhy Own the Hardware (34:39)* Harmoni’s proprietary devices with embedded Linux, RFID, and edge AI* Patented RFID workflows for tracking people, jobs, tools, gauges, and fixtures* The factory-grade indicator light that shows OEE in real time from across a 100,000-square-foot floor* Why edge inference matters: streaming constant data to the cloud is impractical on cost and latencyThe Series A with Bessemer (37:43)* Bessemer’s early scout investment and how the relationship grew* Meeting partner Kent Bennett and what an immediate connection looked like* Closing the $10M Series AHAL and the Context Problem (39:08)* Why putting ChatGPT or Claude on a laptop at every machine doesn’t work: the missing piece is context* Everything an operator would have to feed a model by hand: the part, the program, the measurements, the maintenance history, the sensor data* How HAL delivers that context automatically, at the machine and at the moment of work* HAL in the cloud for managers, and why frontline operators are where it matters mostReindustrialization and the $3 Trillion Installed Base (43:38)* Why autonomous factories and legacy shops both need to exist* The math: greenfield autonomous factories might reach $100B in revenue, leaving about $2.9T made the old way* Why OEMs like Boeing won’t move critical parts like landing gear for a 20% saving, given safety risk and the web of regulated outside vendors* Why only about 10% of a part’s cost is the machining itself* Harmoni’s mission: helping the installed base defend itself against the coming pressure from autonomous factoriesWhy Hardware Is Back in Style (47:50)* “Hardware is hard,” and why that makes it a real moat when software barriers are softer* Hardware only counts as a moat if it is crucial to the product* Why Harmoni thinks of itself as hardware-enabled SaaSWhat Technologists Get Wrong About the Shop Floor (49:16)* Why operators are jaded by software that feels like an ankle bracelet* The test for any tool: how will it change what actually happens on the floor?* Why pretty dashboards without changed behavior shouldn’t be expected to deliver ROIAbout David CaputoDavid Caputo is co-founder of Harmoni, a factory orchestration system bringing proprietary devices and custom software to the shop floor. Previously, he spent nearly a decade in leveraged finance at RBC Capital Markets, then founded RVE Partners to acquire and operate aerospace and defense manufacturers across the middle of America.Connect with David Caputo: linkedin.com/in/david-caputo-52399499Visit the Harmoni website: https://harmoni.io/Connect with Matt Cohen on LinkedIn: https://ca.linkedin.com/in/matt-cohen1Visit the Ripple Ventures website: https://www.rippleventures.com/ This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit tanktalks.substack.com
Welcome back to another episode of Tank Talks: The Rundown! Matt Cohen and John Ruffolo unpack a week of leadership upheaval, takeover proposals, and technology failures. Jensen Huang’s departure puts Nvidia’s role at the center of the AI economy under scrutiny. Québecor’s bid for Telus raises fresh questions about Canadian telecom competition. And a string of technology outages sparks a bigger conversation about how much of everyday life depends on systems we barely understand.For founders, investors, and anyone following Canadian business and technology, Matt and John connect the week’s headlines to the decisions that could shape what comes next.Nvidia after Jensen Huang: Is the AI ecosystem too dependent on one company? (01:42)Jensen Huang’s resignation raises a question that reaches far beyond Nvidia: what happens to the AI industry when its most influential company faces a leadership crisis? Matt and John examine Nvidia’s position at the center of AI infrastructure spending, its relationships with neocloud providers, and the financing arrangements connecting the industry’s biggest players. Nvidia confirmed Huang’s departure and appointed CFO Colette Kress as interim CEO.The governance lesson: How successful companies leave themselves exposed (07:24)When a company’s stock keeps climbing, uncomfortable questions can become easy to postpone. Matt and John discuss founder accountability, board oversight, and the danger of assuming extraordinary business results mean the organization behind them is equally strong.John argues that a complacent board can allow vulnerabilities to accumulate until a crisis makes them impossible to ignore. Matt asks what happens when legal scrutiny, shareholder pressure, and leadership uncertainty arrive at once.Québecor’s Telus bid: Would Canadians pay the price for less competition? (09:21)Québecor’s proposed acquisition of Telus puts Canadian telecom consolidation back in the spotlight. Matt and John examine the proposal discussed in the episode, Telus’s debt burden, and the argument that a larger combined operator could invest more while offering better consumer pricing. Telus has confirmed receiving an unsolicited, conditional proposal and identified significant regulatory and execution risks.Canada, China, and U.S. trade: Is Carney creating leverage? (14:03)Canada’s outreach to China creates an awkward diplomatic balancing act while U.S. trade negotiations continue. Matt questions whether closer ties with Beijing could complicate Canada’s relationship with Washington.John sees a possible strategic purpose. Even agreements with limited immediate economic impact can signal that Canada has other relationships to develop, potentially strengthening its negotiating position.Technology outages and cybersecurity: What happens when the systems stop? (15:18)Disruptions involving workplace software, aviation, banking, and payments lead Matt and John into the episode’s most unsettling question: how resilient are the systems society depends on?They discuss reported outages, the risks of internet-connected devices, and whether AI could make software vulnerabilities easier to find and exploit. Matt also shares a cybersecurity founder’s pushback on the idea of a coordinated cover-up. The conversation does not establish a common cause for the disruptions.From AI infrastructure and corporate governance to Canadian telecom competition and cybersecurity, this episode asks who holds the power, who provides oversight, and what happens when critical systems fail.Connect with John Ruffolo on LinkedIn: https://ca.linkedin.com/in/joruffoloConnect with Matt Cohen on LinkedIn: https://ca.linkedin.com/in/matt-cohen1Visit the Ripple Ventures website: https://www.rippleventures.com/ This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit tanktalks.substack.com
Neil Cawse spent years selling vehicle tracking to executives who told him they would never spy on their employees. Today, Geotab connects roughly six million vehicles and assets, serves more than 100,000 customers, and has grown past $1 billion in revenue without outside funding. In this episode of Tank Talks, host Matt Cohen sits down with Geotab’s founder and CEO to explore how he built one of Canada’s largest private technology companies, survived a cash crunch, and stayed patient until the fleet telematics market caught up.Now Neil is working through another major shift: making Geotab an AI-native company. He shares what happened when the business spent $800,000 on AI tokens in 30 days, why adoption requires more than giving employees access to tools, and how AI is changing the value of software. They discuss the competitive advantage of Geotab’s connected vehicle data, the hardware and customer relationships that are harder to replicate, and why Neil believes generalists who know how to use AI will have an edge.The conversation also covers the growth opportunity he wishes he had pursued more aggressively, the tradeoffs of staying private, acquisitions in fleet management and asset tracking, and what autonomous vehicles mean for Geotab’s future. For founders and operators navigating bootstrapping, business growth, and AI adoption, Neil offers a candid account of the decisions that worked, the mistakes he still thinks about, and the changes he’s making now.The opportunity behind GPS and vehicle telematics (04:47)* How the possibilities of GPS led Neil to explore vehicle tracking, fleet productivity, and safety* His approach to entrepreneurship: connecting emerging technology with practical business problems* Why he nearly returned to university to study AI instead of pursuing GeotabSurviving when the market isn’t ready (08:24)* Why Geotab spent years in a niche market before customers became comfortable with vehicle tracking* The resistance Neil encountered from executives who saw telematics as employee surveillance* Why he believes taking outside funding too early could have put the business under pressure it couldn’t surviveThe cash crunch that tested Geotab’s bootstrap strategy (12:34)* Investing an initial $500,000 and approaching the point where another funding decision was necessary* Bringing the team together to assess whether customer opportunities justified continuing* How low costs, modest salaries, and progress with UPS helped Geotab reach firmer financial groundThe ELD opportunity he wishes he had pursued harder (16:14)* How the U.S. electronic logging device mandate created a major opening in fleet technology* Why Neil considers Geotab’s conservative response a missed opportunity* The challenge of reaching individual truck owners compared with selling to large fleetsWhy Neil thinks generalists will win with AI (22:05)* How AI is forcing him to rethink established engineering practices and move faster* Why understanding how the pieces of a business fit together becomes more valuable as AI handles more execution* The practical importance of learning tools such as Claude Code and Codex and applying them to real workflowsBuilding an AI-native company and spending $800,000 on tokens (25:34)* Working through employees’ concerns about job security, reliability, and changing familiar habits* Giving AI access to business systems, then supporting adoption through training and useful examples* Why a rising token bill pushed the discussion toward measurable outputs* Neil’s goal of doing twice as much with the people Geotab already employsStaying private while competing with public companies (31:05)* The freedom to make long-term investments without managing the business around quarterly results* The visibility and relationship advantages that publicly traded competitors can gain* Why Neil wants to preserve the option of an IPO while continuing to value private ownershipWhat happens to telematics when the driver disappears? (35:22)* The potential role of vehicle data in autonomous vehicle accident reconstruction and accountability* Why Neil sees long-haul trucking as an important early market for autonomy* How maintenance, routing, and electric vehicle charging remain relevant as driver-focused services changeTurning Geotab’s data into an autonomy business (38:54)* Why Neil sees commercial rollout timing as the main factor in demand from autonomous vehicle operators* How Geotab’s Altitude platform could support fleets moving beyond pilots* The opportunity to help operators manage larger numbers of vehicles over longer distancesWhat Canada needs to change about entrepreneurship (41:45)* Why Neil believes greater competitiveness requires more calculated risk-taking* Teaching children to experiment, learn from failure, and consider entrepreneurship* His argument for a stronger appreciation of entrepreneurs’ role in creating jobs and economic opportunityAbout Neil CawseNeil Cawse is the founder and CEO of Geotab, a Canadian connected transportation and fleet telematics company based in Oakville, Ontario. An electrical engineer originally from South Africa, he built Geotab without outside funding into a business serving more than 100,000 customers. Its technology helps organizations improve fleet safety, vehicle maintenance, productivity, and operational efficiency through connected vehicle data and analytics.Connect with Neil Cawse on LinkedIn: ​​https://www.linkedin.com/in/neilcawse/Visit the Geotab website: https://www.geotab.com/Connect with Matt Cohen on LinkedIn: https://ca.linkedin.com/in/matt-cohen1Visit the Ripple Ventures website: https://www.rippleventures.com/ This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit tanktalks.substack.com
Welcome back to another news roundup on Tank Talks: The Rundown. The federal budget just dropped, and in this emergency episode, Matt Cohen and John Ruffolo cut through the noise to focus on what matters for tech and venture capital. The verdict: the Canada Growth Incentive hit the mark.Both Matt and John spent the past year pushing Ottawa to close the gap with U.S. capital gains rules. Matt’s Build Canada policy memo, Reward the Risk Takers, called for exactly that. The ask was two things: a meaningful capital gains exemption for investing in early-stage Canadian companies, and a rollover to keep those gains in the ecosystem. The government delivered both.For founders, early employees, and investors, this episode breaks down who qualifies, what the fine print says, and why nobody should be calling their accountant just yet.Budget 2026: The Ask Was Two Things. Ottawa Delivered Both (01:42)Everyone has become an economist overnight, with reactions ranging from “great news” to “worst budget ever.” Matt and John set the politics aside and zero in on the innovation economy.John explains the problem they set out to fix: capital formation for IP-based technology businesses. U.S. Qualified Small Business Stock rules, commonly known as QSBS, have created what John calls a giant sucking sound, pulling Canadian founders south because their investors get a better deal there. Meanwhile, resource and large CapEx players got major deductions in past budgets, while the new economy got nothing.Inside the Canada Growth Incentive (04:26)Canada has 33 small businesses per thousand people. The U.S. has 124. The Canada Growth Incentive is designed to change how small businesses get created and funded here.Matt walks through the details:* Up to $15 million CAD in capital gains exemptions per individual, per business, not a single lifetime exemption on one exit* Applies to Canadian-controlled private corporations (CCPCs) with $100 million or less in assets at the time of issuance* Minimum hold increases to three years, up from two* The business must be primarily carried on in Canada* Carve-outs for real estate, traditional consulting and services businesses, and potentially some hybrid fintech modelsJohn is clear about the intent: this was never meant for law firms, accounting practices, or lawn mowing businesses. It is a direct competitive response to the U.S., aimed at scalable, IP-based companies. Is it a coincidence that the seven largest companies in the world by market cap are U.S. technology companies, all built on early-stage venture dollars? Canada needs to play in that game.The Rollover: Keeping Exit Dollars in Canada (06:47)Under the old lifetime capital gains exemption, a founder or investor got one shot. Once it was used, there was little reason to put the next dollar back into Canadian startups instead of following U.S. incentives.The new rollover changes that. Proceeds can be reinvested into another qualifying Canadian small business within the year of sale plus three years, extended to five years when a SAFE is involved. John calls the timeline unbelievable, having expected one year. The measures are forward-looking, running through 2037.Open questions remain, including whether a fintech counts as financial services or technology.Take It Easy: This Is Not Law Yet (08:21)For anyone already dialing their accountant: slow down. The budget was tabled on November 5 as a fiscal plan with notices of ways and means motions. It still needs to pass through a budget implementation act.Two things to know now:* It is not retroactive. It applies to newly issued shares, not investments already on your books.* Options do not count. Employees need to own shares, not hold unexercised options. Matt saw this firsthand when Turnstyle was sold: many employees had meaningful ownership through options they never exercised, so the hold period never started.John’s advice for early employees with low strike prices: exercise the option, and the rules start applying to you. Cashless exercise is another route for those without the cash.The Canada Strong Fund: A $25B Sovereign Wealth Fund (10:47)The big headline was the Canada Strong Fund, a $25 billion sovereign wealth fund to be set up as a Crown corporation. It will take minority co-investments on commercial terms, make selective direct investments, and offer retail notes to Canadians.John is glad it is launching, but questions the timing on retail notes, which work like the old Canada Savings Bonds: a guaranteed minimum return with upside. His advice: get the fund running and work out the kinks first. The fund does not even have a CEO yet.Matt adds that without a surplus, the fund is not investing off a strong balance sheet. Its real value may be as an anchor for foreign direct investment, putting a stake in the ground that brings global capital in alongside it.The budget also included the previously announced venture capital pools and $675 million for growth-stage investment through BDC.The BDC-EDC Merger: A Head-Scratcher on Timing (13:42)BDC and EDC will be merged into a single federal development bank, with no timeline yet and existing products continuing in the meantime.John explains how the two drifted into overlap: BDC focused on early stage, EDC moved into growth, and sectors like defense saw both pile in. Consolidation could reduce founder confusion and create cost synergies, but the real driver is unclear. Matt questions adding a merger on top of BDC’s already heavy mandates this year. Both expect it to take a while.Execution Is Now on the Private Sector (16:02)The budget also gives Minister Evan Solomon’s team plenty to work on, including a new super regulator and a Canadian Digital Safety Commission. Add real-time payment rails, and much of what the tech sector has asked for over the last decade is finally on paper.John’s closing message: is any of this perfect? No. But the tech sector asked for two specific things and got both, which almost never happens. Take the win. Ottawa listened. Now it’s up to the private sector to execute. It’s time to build.Connect with John Ruffolo on LinkedIn: https://ca.linkedin.com/in/joruffoloConnect with Matt Cohen on LinkedIn: https://ca.linkedin.com/in/matt-cohen1Visit the Ripple Ventures website: https://www.rippleventures.com/ This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit tanktalks.substack.com
In this episode of Tank Talks, host Matt Cohen welcomes back Peter Walker, Head of Insights at OpenRouter and formerly of Carta, where his charts on cap tables and fundraising became required reading for founders and VCs. Peter now sits on one of the largest real-world data sets in AI, with roughly 600 trillion tokens a month flowing through a single API that gives access to 500+ models. He explains why OpenRouter is as much about routing between inference providers as between models, and why the same model can feel brilliant or terrible depending on who serves you the token.They also dig into how fast the market has splintered beyond Claude and GPT, why agentic usage was the turning point, and why most teams don’t need the most expensive token for every task. Peter shares his view on open-weight models and the geopolitics behind them, why the AI “bubble” looks different from the dot-com fiber glut, and what stood out to him in Anthropic’s S1, from the non-cancellable compute commitments to the eye-popping growth and net dollar retention. He also covers neo-clouds, fine-tuning versus distillation, why he thinks “model nerfing” is a red herring, and why cultural change, not technology, will slow AI’s spread through the broader economy.Whether you’re building on top of the AI stack, funding it, or just trying to make sense of it, this conversation offers a data-driven view from someone who watches the whole market.From Carta to OpenRouter (03:02)* Peter’s path from a data-visualization nerd at Carta to head of insights at OpenRouter* Why he wanted to get closer to AI, and the FOMO that pulled him there* Why OpenRouter’s cross-lab view is unique: labs only see their own data, while OpenRouter sees which models, tasks, and providers are gaining traction* Joining 10 days before the Stripe acquisition was announcedWhat OpenRouter Actually Does (04:50)* One API key for 500+ models, instead of being “a Claude shop” or “an OpenAI shop”* Why the more important layer is inference provider routing: the same open-weight model can be served by hundreds of providers with very different cost, latency, throughput, and reliability* The marketplace model: more models attract more users, which attract more providers, with a service fee on top* The founding team’s commitment to making as much data public as possible* Why Stripe and OpenRouter fit: both abstract complexity (payments and AI), and AI is now the first or second fastest-growing line item for CFOsFragmentation and the Agentic Turning Point (10:00)* Why the market is still in early innings, from ChatGPT’s launch in late 2022 to today’s diversity of models* The shift in February 2026 from chatting with AI to agents doing the work, and why it kicked off multi-model usage* Matching models to tasks: security reviews, customer service, and why you may not need a frontier model to review your next email* Why this market, with commodity-like supply and complex demand, has no real precedentBad Tokens, Good Tokens (12:54)* Why the inference layer is more complex than the model layer* Closed models are served by a handful of places, while open models have hundreds of providers, some excellent and some “super shady”* Speed differences of up to 20x between providers serving the same model* Why people disagree so much about the same model: often it’s the token they were served, not the model* The reputational risk of putting a slow or unreliable provider behind a customer-facing appSubscriptions vs. API Pricing (15:35)* A carnival-token analogy for how tokens work* Why $100-a-month subscriptions are still fantastic deals, and why API bills can become “eye-watering”* Why the best models are still worth it for many use cases, but not for all of them* The geopolitical layer: open-weight models skew Chinese, closed-weight models skew AmericanOpen Weights vs. the Frontier (18:41)* Why OpenRouter believes in a multiplicity of winners rather than one open-or-closed champion* New model categories nobody had envisioned weeks ago, such as fast, cheap decision models for classification* Why not everything will happen inside OpenAI or AnthropicIs AI a Bubble? (21:42)* Why this isn’t the dot-com fiber glut: everyone is compute constrained, and more GPUs would be used* How supply and demand can still get out of whack* Why there’s no world where we go back to a pre-AI economyReading Anthropic’s S1 (22:37)* The scale of the multi-year, largely non-cancellable compute commitments* Why Peter sees Anthropic as highly profitable on inference, with the open question being whether it can fulfill its commitments* 14x growth last year from a base of roughly $780 million, at a company only three years old* The reported 900% net dollar retention, and why it’s remarkable even if it doesn’t repeat* Why using more Anthropic and adding open-weight models to the portfolio aren’t mutually exclusiveNeo-Clouds and the GPU Gold Rush (26:24)* Where the next token gets served from when most of Anthropic’s compute is with the hyperscalers* 50+ inference providers apply to join OpenRouter each week, and very few are accepted* What a neo-cloud really is: capital-intensive GPU access plus software that runs models faster and more reliably* Why it’s hard: capital markets, energy supply, infrastructure teams, 100% uptime, and deal-making skills that SaaS never needed* Neo-labs vs. neo-clouds, and why many labs will transition into cloudsFine-Tuning, Distillation, and Model “Nerfing” (31:38)* Fine-tuning: how legal AI companies turn a strong base model plus proprietary data into a specialized product* Distillation: why closed labs forbid it, why the open-weight rules are murkier, and how labs distill their own models into new variants* Why “model degradation” is a red herring: slower or less reliable tokens at the inference layer and the self-cannibalizing release cycle are more likely explanationsModels vs. Harnesses (36:00)* How much product and harness matter now that raw intelligence is less differentiating for some use cases* Why Peter thinks model intelligence still matters most for the top labs, and the product is everything for application-layer companies* Personal agents as the first time many people experience powerful AI without knowing which model is behind itThe Cultural Gap in AI Adoption (40:56)* Why the biggest divide is generational, including the age of the company* Why diffusion beyond San Francisco and New York will likely disappoint, because it’s a person-to-person problem, not a technological one* How doom-and-gloom messaging causes people to disengage* Why everyone now has “the smartest EA in the world” in their pocketGrieving What You Were Great At (45:49)* Peter’s own experience letting AI build the charts he spent a decade mastering* Accepting the inevitability of the change and moving your expertise to the strategic layer: asking the right questions and providing the right framing* Finding new things to do that the technology enables rather than removesAbout Peter WalkerPeter Walker is Head of Insights at OpenRouter, where he analyzes usage data across hundreds of models and inference providers. Previously, he spent five years at Carta turning cap table, valuation, and fundraising data into widely read charts and research.Connect with Peter Walker: https://www.linkedin.com/in/peterjameswalkerVisit the OpenRouter website: https://openrouter.ai/Connect with Matt Cohen on LinkedIn: https://ca.linkedin.com/in/matt-cohen1Visit the Ripple Ventures website: https://www.rippleventures.com/ This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit tanktalks.substack.com
loading
Comments