Code and Cobblestones

Every day, thousands of trading cards, coins and sneakers get listed online by people guessing what they're worth. Some are mislabeled and some are fake. The experts who could tell the difference are backed up by millions of cards, and grading one can take nine months and $50. So why can't the biggest AI models fix this? That's what this episode is about.
In episode 12 of Code and Cobblestones, Derek Bugley, CEO of Vardera, joins us, Matt Crane and Will Lehmann, to talk about building what he calls the intelligence layer for collectible commerce. Vardera only needs the front and back of a card. From that it can identify the card, grade it, price it and flag whether it's fake, all in about 300 milliseconds.
The idea started at home. Derek and his co-founders, Michael and Ben, lost family members around the same time and ended up handling estates full of collectibles. (His dad collected sports memorabilia.) They watched appraisers walk through, take photos and get the values wildly wrong. Then they started talking to the biggest marketplaces in the world. What looked like a personal problem turned out to be a multi-billion-dollar one.
We got into:
Why Vardera tested frontier models first, spent a lot of money on mixed results, and now builds everything from scratch
The last-mile problem: delivering 99.5% accuracy and getting asked for more
Why half the business is data labeling, and why "the data sucks" at companies that claim to have the best data
Skipping the consumer app ("the startup graveyard") to sell to marketplaces, plus an open API built for AI agents
A blind head-to-head against human graders, where Vardera scored in the high 90s and caught cards that had already been slabbed with the wrong grade
Grading backlogs of 20 to 30 million cards, and why nobody's training to be a numismatist anymore
Live commerce, Whatnot, and a fisherman in Seattle selling same-day salmon on a livestream
Collectibles as an asset class, and why Derek wants Vardera to be the Stripe of this market
Why the next generation of heirs will sell the art and buy cards, sneakers and sealed Nintendo games
Rapid fire: an $80 Castle Grayskull vs. one that sold for $240K, NFTs "valuable or BS?" (his answer: "Yes"), and why Boston doesn't need an 11th steakhouse in the Seaport
Mentioned in this episode:
Vardera's breakdown of how its grading model compares to frontier AI models: https://www.vardera.com/blog/we-built-the-best-trading-card-identification-and-grading-model-in-the-world.-here-are-the-numbers
 
TIMESTAMPS: 00:00 Meet Vardera CEO Derek Bugley 00:31 AI for Collectible Authentication 01:35 OpenAI Buys Glass Imaging 04:43 Meta Glasses and Computer Vision 06:56 Collectibles API for AI Agents 10:03 Custom AI vs Frontier Models 14:05 Vardera's Founding Story 16:16 Card Grading Backlogs at Scale 19:03 Choosing Collectible Categories 24:09 AI Grading vs Human Graders 26:21 Fraud in the Collectibles Market 27:04 Working With Auction Houses 28:21 Why Human Experts Don't Scale 31:23 How Vardera's AI Works 34:23 Live Commerce and Whatnot 37:31 The Stripe of Collectibles 41:34 Building a Boston AI Startup 43:35 Rapid Fire 49:25 Advice for Founders
 
Guest: Derek Bugley, Co-Founder and CEO of Vardera
https://www.linkedin.com/in/derekbugley/ https://www.vardera.com/ 
 
Hosts: 
 
Matt Crane, Founder of MGMT Boston https://mgmtboston.com/ 
https://www.linkedin.com/in/matthewhcrane/ 
 
Will Lehmann Founder of Step Function VC https://stepfunction.vc/ 
https://www.linkedin.com/in/wlehmann/ 
Watch/listen/Subscribe: 
Spotify: https://open.spotify.com/show/033HueFNlE1LPnEFPvVscs?si=e2351085b40b4334 
Apple Podcasts: https://podcasts.apple.com/us/podcast/code-and-cobblestones/id6785977680 
Youtube: https://www.youtube.com/@MGMTBoston 
 
Code and Cobblestones is a love letter to Boston tech founders. New episodes spotlight the people building the city's next generation of companies.
 
▶ Subscribe for the rest of the tour. New episodes spotlight the founders and builders putting Boston on the map.
 
Produced in partnership with OBEY Creative, building podcast-driven content engines for B2B founders. One great conversation per week becomes 15-20 assets: clips, LinkedIn posts, blog content, and newsletter material. Learn more at https://obeycreative.com/

Most companies think they lose customers at renewal. They actually lose them long before that, during onboarding. So why do so many still run it on spreadsheets and email threads? And what changes when AI agents start doing the work? That's what this episode is about.
 
In episode 11 of Code and Cobblestones, Paul Holder, co-founder and CEO of OnRamp, joins us, Matt Crane and Will Lehmann, to talk about the part of the customer journey he calls the most important one. By his count, 90% of the time onboarding is the reason a customer renews or walks.
 
He learned that the hard way. Leading onboarding at VTS, back when the answer to scale was just hiring more people. Running post-sales at Troops and seeing the same frustrated customers. Then a question for his co-founder, Ross: are we just bad at our jobs, or is everyone feeling this? Turns out everyone was. They started OnRamp in the middle of COVID, fully remote, and missed the 2021 valuation frenzy entirely. He's grateful for that now.
 
We got into:
Why Paul thinks autonomous AI agents can finally take real action in B2B, and why that answer wasn't true six months ago
PLG vs. enterprise onboarding, and what happened when Slack tried to roll out to IBM (he had a word for it we can't print here)
"Making progress between the calls," and why most onboardings stall waiting for the next meeting
Rebuilding a SaaS company as AI native without throwing out the foundation. No AI for the sake of AI.
New pricing coming soon (breaking news, live on the show), and the catch-22 of charging for every AI action
Why Onramp stopped selling to 30-person startups, and how cutting onboarding drop-off by 10% turns into a $3M problem for one customer
Raising $27M with a 20, 30, 40-year view, and why a sexy headline valuation was never the goal
His advice after a Series A: don't hire ahead of growth. Wait until the team is bursting at the seams.
Two CEO careers, two young kids, and a paper calendar in the kitchen that beats most of the tech
 
TIMESTAMPS:00:00 Intro: Paul Holder, OnRamp CEO
00:41 AI Personal Assistants: Hype Check
01:34 AI Adoption in B2B
02:47 Autonomous AI Agents in B2B
04:28 Dreamforce and AI Interfaces
06:58 PLG vs Enterprise Onboarding
09:45 When AI Agents Onboard AI Agents
10:49 Onboarding Beyond Spreadsheets
11:39 Onramp Origin Story
15:51 From SaaS to AI Native
19:00 How to Price AI Features
20:49 Finding Your ICP
23:57 Who Onramp Is Built For
24:46 Should AI Companies Sell Services?
25:12 From Product to Platform
27:19 Customer Obsession Culture
29:26 High Standards, No Gimmicks
31:02 Fundraising and Cap Table Strategy
33:35 Founder Lesson: Trust Your Gut
34:58 Balancing Family and CEO Life
37:04 How a CEO Spends the Week
38:22 Rapid Fire: Mentors and Series A Hiring
42:49 Tools for Busy Founder Parents
45:23 Why Now Is the Time to Build
Guest: Paul Holder, Co-founder & CEO of OnRamp
https://www.linkedin.com/in/paulcholder/ https://onramp.us/ 
 
Hosts: 
 
Matt Crane, Founder of MGMT Boston https://mgmtboston.com/ 
https://www.linkedin.com/in/matthewhcrane/ 
 
Will Lehmann Founder of Step Function VC https://stepfunction.vc/ 
https://www.linkedin.com/in/wlehmann/ 
 
Watch/listen/Subscribe: 
Spotify: https://open.spotify.com/show/033HueFNlE1LPnEFPvVscs?si=e2351085b40b4334 
Apple Podcasts: https://podcasts.apple.com/us/podcast/code-and-cobblestones/id6785977680 
Youtube: https://www.youtube.com/@MGMTBoston 
 
Code and Cobblestones is a love letter to Boston tech founders. New episodes spotlight the people building the city's next generation of companies.
 
▶ Subscribe for the rest of the tour. New episodes spotlight the founders and builders putting Boston on the map.
Produced in partnership with OBEY Creative, building podcast-driven content engines for B2B founders. One great conversation per week becomes 15-20 assets: clips, LinkedIn posts, blog content, and newsletter material. Learn more at https://obeycreative.com/

Most founders think a killer pitch deck is what gets them funded. It isn't. The best early-stage investors are writing checks before there's a product, before there's revenue, sometimes before there's even a company name. So what actually convinces them? That's what this episode is about.
In episode 10 of Code and Cobblestones, Ellen Chisa joins us, Matt Crane and Will Lehmann, to walk through what it takes to get funded at the earliest stage there is: inception. She's a partner at Boldstart Ventures, where they only write checks before a product exists, and she's been doing it for six years.
She didn't take a straight line to venture capital. She dropped out of college for a year to start a company with five friends. She built the first versions of Office Mobile at Microsoft, then worked at Kickstarter and at Lola. After that she co-founded Dark, a programming language company, with Paul Biggar, who started CircleCI before that. Dark didn't end the way either of them planned. She walked away, and Boldstart brought her in instead.
What we talked about:
Why Ellen doesn't need a pitch deck, and what she actually asks for instead: a ten-year vision you can defend, and a plan for next week
The "AB flip-flop," the pattern she watches kill startups: go end-user first, flip to enterprise, flip back, never actually commit to a lane
Why she thinks every founder has a hard ceiling of 50, maybe 70, good years in them, and what that means for knowing when to walk away from something that isn't working
Anthropic's new 25,000-square-foot office in Kendall Square, and why she thinks the real draw isn't PR, it's the talent pool at MIT and Harvard
Bending Spoons buying Miro, and what she says it actually reveals about which startups have the growth to back up their old valuations, and which don't
Why she thinks embodied AI is still underrated next to software agents, twelve months out (her prediction, not ours)
 
TIMESTAMPS:00:00 Ellen Chisa Joins the Show
01:07 AI Safety Debate and P(doom)
05:02 Anthropic Opens a Boston Office
07:25 Miro Sold, SaaS Valuations Reset
08:40 Ellen Chisa's Path to Boldstart
12:57 Building Dark: A Founder's Story
15:50 Why Ellen Walked Away From Dark
17:48 Becoming a VC at Boldstart
20:54 How Founders Should Use VCs
22:48 Ellen's Next Big Bet
25:05 Inception Stage Investing Explained
27:23 Boston Startups: Rapid Fire
32:00 Advice for First-Time Founders
33:48 Where to Find Ellen Chisa
Guest: Ellen Chisa, Partner at Boldstart Ventures 
https://www.linkedin.com/in/ellenchisa/ 
 
Hosts: 
Matt Crane, Founder of MGMT Boston https://mgmtboston.com/ 
https://www.linkedin.com/in/matthewhcrane/ 
 
Will Lehmann Founder of Step Function VC https://stepfunction.vc/ 
https://www.linkedin.com/in/wlehmann/ 
 
Watch/listen/Subscribe: 
Spotify: https://open.spotify.com/show/033HueFNlE1LPnEFPvVscs?si=e2351085b40b4334 
Apple Podcasts: https://podcasts.apple.com/us/podcast/code-and-cobblestones/id6785977680 
Youtube: https://www.youtube.com/@MGMTBoston 
 
Code and Cobblestones is a love letter to Boston tech founders. New episodes spotlight the people building the city's next generation of companies.
 
▶ Subscribe for the rest of the tour. New episodes spotlight the founders and builders putting Boston on the map.
Produced in partnership with OBEY Creative, building podcast-driven content engines for B2B founders. One great conversation per week becomes 15-20 assets: clips, LinkedIn posts, blog content, and newsletter material. Learn more at https://obeycreative.com/

Hackers already have AI. So do the people trying to stop them. Who wins that race? And how fast is it actually moving? That's what this episode is about.
 
In episode 9 of Code and Cobblestones, Lior joins us, Matt Crane and Will Lehmann, to walk through what it looks like to build the defender side of an arms race most people don't know is already happening. Hackers already have AI writing and running their attacks. Almost nobody has AI running the defense with the same speed.
 
He didn't come at this from a typical tech background. Unit 8200 at 18. Years with the Mossad that don't show up on LinkedIn. Then Cybereason, one of the first EDR companies, built over 12 years. Now 7AI, built with the same co-founder, Yonatan, after a conversation with OpenAI's former head of product convinced him the ground was shifting under the entire industry.
We got into:
Why Lior thinks we're racing toward a moment when AI can plan, execute, and repeat cyberattacks entirely on its own, and why he still believes defenders can win that race
Open weight AI models, which he called a gift with sharp edges, and why he thinks the US should lean into them instead of running from them
Autonomous code generation, and why that ship has sailed whether security teams are ready or not (his words, not ours)
What 7AI actually does day to day, and why Lior says "AI SOC" is already commodity. Too little, maybe too late, if that's still your whole pitch.
The three-star Michelin culture 7AI is building internally, and why some of their strongest hires never worked in tech a day in their life before joining
How 7AI runs as a flat, fast organization with almost no recurring meetings, and why Lior says that only works because of how they hire
 
TIMESTAMPS:00:00 Lior Div, 7AI CEO
01:34 AI M&A: Stripe and NVIDIA
03:47 AI Cyberattacks Are Rising
06:09 Open Weight Models Debate
08:27 Risks of AI Code Generation
10:35 Lior's Path: Unit 8200 to Mossad
13:29 Why He Founded 7AI
15:34 What 7AI Actually Does
17:02 Beyond AI SOC
19:38 Building an AI-Native Company
24:10 Hiring for a Builder Culture
28:35 The Michelin Hiring Approach
32:46 Raising Capital for 7AI
35:00 Why Build in Boston
37:10 Rapid Fire
 
Guest: Lior Div,  Co-founder and CEO of 7AI https://7ai.com/ https://www.linkedin.com/in/liordiv/ 
 
Hosts: 
 
Matt Crane, Founder of MGMT Boston https://mgmtboston.com/ 
https://www.linkedin.com/in/matthewhcrane/ 
 
Will Lehmann Founder of Step Function VC https://stepfunction.vc/ 
https://www.linkedin.com/in/wlehmann/ 
Watch/listen/Subscribe: 
Spotify: https://open.spotify.com/show/033HueFNlE1LPnEFPvVscs?si=e2351085b40b4334 
Apple Podcasts: https://podcasts.apple.com/us/podcast/code-and-cobblestones/id6785977680 
Youtube: https://www.youtube.com/@MGMTBoston 
 
Code and Cobblestones is a love letter to Boston tech founders. New episodes spotlight the people building the city's next generation of companies.
 
▶ Subscribe for the rest of the tour. New episodes spotlight the founders and builders putting Boston on the map.
Produced in partnership with OBEY Creative, building podcast-driven content engines for B2B founders. One great conversation per week becomes 15-20 assets: clips, LinkedIn posts, blog content, and newsletter material. Learn more at https://obeycreative.com/

Every self-driving car has cameras or a lidar system. It knows what's around it before it decides where to go. Now picture a factory built the same way, except nobody ever built the camera or the lidar. That's process manufacturing today: the reactors, pipes, and stainless steel equipment behind your protein bar, your prescription, the chips in your laptop. Half of everything the world manufactures runs through equipment like this, and almost nobody measures what's actually happening inside it in real time.
 
That's the gap Annie Lu and her brother David set out to close with Laminar.
 
In episode 8 of Code and Cobblestones, Annie joins us, Matt Crane and Will Lehmann, to walk through what physical AI looks like when it's not a robotics story. No sensing the room, deciding, moving an arm. Here the thing being sensed is a chemical reaction, not a room, so Laminar had to build its own version of that camera and lidar from scratch: proprietary sensors first, then closed foundation models trained on the chemistry data, then a direct line into the factory's own control systems.
 
They didn't start with the biggest possible use case. They started with cleaning. Every plant has to shut down between batches and wash everything out, so one product doesn't contaminate the next. Right now, almost every factory in the world runs that cleaning on a fixed schedule that never adjusts, and burns water, energy, chemicals, and downtime doing it. Laminar figured out how to make that process dynamic instead of fixed, and calls it the wedge that earns them the right to expand into everything else: fermentation, filtration, distillation, and beyond. It's why they're now running across six continents, inside seven of the top ten food and beverage companies in the world.
 
What you'll learn:
Why Annie treats robotics as the wrong template for physical AI. Chemistry intelligence, not spatial intelligence, is the actual gap
How Laminar's stack works end to end: sensors on the equipment, foundation models reading the chemistry, automation acting on it, with zero human in the loop
Why clean-in-place became the wedge instead of something bigger or flashier, and how one use case earns the right to expand across a whole plant
The real reason legacy manufacturing tech hasn't changed in 60 years, per Annie. It's not that factories resist AI. Nobody built the camera and lidar for chemistry
What happened when a competitor told a room of plant managers their model was a black box they couldn't explain, and watched the room turn
Her advice for founders on choosing investors: find the ones who already match your DNA, before you ever pitch them
Why she thinks Boston's biggest advantage is culture, not capital, and what that means for a physical AI company scaling globally
 
TIMESTAMPS:
00:00 Welcome 
01:05 Why Physical AI Now
03:38 Laminar and Process Manufacturing
06:24 The Chemistry Intelligence Stack
09:51 Laminar's Origin Story
12:32 First Principles on the Shop Floor
17:40 Co-Founding With Her Brother
20:32 Winning Trust From Manufacturers
26:57 The Clean-in-Place Wedge
31:57 Advice for Physical AI Founders
35:24 Fundraising and Investor Fit
38:46 Rapid Fire With Annie Lu
40:07 Boston's Manufacturing Culture
42:17 How to Work With Laminar
 
Guest: Annie Lu, Co-founder, CEO at Laminar 
 
Hosts: 
 
Matt Crane, Founder of MGMT Boston https://mgmtboston.com/ 
https://www.linkedin.com/in/matthewhcrane/ 
 
Will Lehmann Founder of Step Function VC https://stepfunction.vc/ 
https://www.linkedin.com/in/wlehmann/ 
Watch/listen/Subscribe: 
Spotify: https://open.spotify.com/show/033HueFNlE1LPnEFPvVscs?si=e2351085b40b4334 
Apple Podcasts: https://podcasts.apple.com/us/podcast/code-and-cobblestones/id6785977680 
 
Code and Cobblestones is a love letter to Boston tech founders. New episodes spotlight the people building the city's next generation of companies.
 
▶ Subscribe for the rest of the tour. New episodes spotlight the founders and builders putting Boston on the map.
 
Produced in partnership with OBEY Creative, building podcast-driven content engines for B2B founders. One great conversation per week becomes 15-20 assets: clips, LinkedIn posts, blog content, and newsletter material. Learn more at https://obeycreative.com/

Shannon Kay studied industrial engineering at Clemson, spent a summer taste-testing soup at Campbell's, cut her teeth at Deloitte, ran supply chain analytics at Wayfair, and left Harvard Business School with an "incomplete" on her LinkedIn to chase an idea that wasn't even the right idea yet. She co-founded Topline Pro anyway, and now runs it as COO.
In episode 7 of Code and Cobblestones, Shannon joins hosts Matt Crane and Will Lehmann to talk about what it takes to build AI for the 90% of small businesses that don't have a tech team, don't sit at a desk, and don't have time to learn point-and-click software.
The conversation runs from a stat that reframes the whole AI conversation (of the 2.6 million home service businesses in the US, only the top 10% have anything resembling a tech stack) to why "trust but verify" matters more than ever after watching an AI-driven hedge fund implode, to the customer who found out their AI assistant was quietly qualifying leads for them while they were out to dinner.
What you'll learn:
Why Shannon built Topline Pro's Sidekick around "get work, win work, get paid" instead of the usual SaaS feature list
How a COVID-era side project called Dwelling turned into ProFo, then Topline Pro, after one wrong pivot pointed her toward the right one
Why she'd rather a customer never open the app at all, and what "done for you" means when your users are on a ladder, not at a desk
The real risk with AI inside a company, per Shannon: not inaccuracy, but rewarding the wrong behavior, and how Topline Pro's "AI-native employee of the month" tries to fix that
What Airtable's $1.2B exit, an AI hedge fund's collapse, and the Agency AI–Klaviyo deal have in common as lessons for founders right now
Her advice for founders building in "unsexy" markets: find the investors who already get it before you ever pitch them
Why incrementalism earns trust with some customers, and showing the whole platform earns it faster with others
 
TIMESTAMPS
00:00 Shannon Kay, Topline Pro Co-Founder Joins the Show
02:19 Airtable's $1.2B Acquisition Explained
05:07 AI Hedge Fund Collapse: Startup Risk Lessons
08:57 Klaviyo Buys Agency AI: Founder Obligations
12:45 Shannon Kay's Career Path: Clemson to Harvard
15:46 Founding Topline Pro: From Dwelling to AI Startup
20:03 Inside Sidekick: AI for Home Service Businesses
22:17 Designing AI Products for Non-Tech Small Businesses
26:40 Building Customer Trust Without Lock-In Contracts
28:54 Raising a $27M Series B: Fundraising Advice
34:01 Rapid Fire: AI Hiring and Boston Startups
39:51 Final Takeaway and Episode Wrap-Up
Guest: Shannon Kay, Co-founder, COO at Topline Pro https://www.toplinepro.com/ https://www.linkedin.com/in/shannonekay/ 
 
Hosts: Matt Crane, Founder of MGMT Boston https://mgmtboston.com/ 
https://www.linkedin.com/in/matthewhcrane/ 
and Will Lehmann Founder of Step Function VC https://stepfunction.vc/ 
https://www.linkedin.com/in/wlehmann/ 
Watch/listen/Subscribe: 
Spotify: https://open.spotify.com/show/033HueFNlE1LPnEFPvVscs?si=e2351085b40b4334 
Apple Podcasts: https://podcasts.apple.com/us/podcast/code-and-cobblestones/id6785977680 
Produced in partnership with OBEY Creative, building podcast-driven authority content engines for B2B founders. Learn more at https://obeycreative.com
 
Code and Cobblestones is a love letter to Boston tech founders. New episodes spotlight the people building the city's next generation of companies.
 
▶ Subscribe for the rest of the tour. New episodes spotlight the founders and builders putting Boston on the map.
Produced in partnership with OBEY Creative, building podcast-driven content engines for B2B founders. One great conversation per week becomes 15-20 assets: clips, LinkedIn posts, blog content, and newsletter material. Learn more at https://obeycreative.com/

Subconscious co-founder and CEO Jack O'Brien makes the case that the ceiling on AI isn't capability, it's cost, and that the next wave belongs to agents that run for hours, not chatbots that answer in seconds. Matt Crane (MGMT Boston) and Will Lehmann (Step Function) get into open weights, edge inference, and why Jack is betting on Boston to build it.
What you'll learn
    Why cost, not capability, is the real ceiling on AI agents
    What "agent abundance" means and why teams get hooked once they have it
    How Subconscious compresses agent context on the GPU (one customer: 95% compressed away, agents never noticed)
    The case for American open weights, and why Jack calls the closed-model argument self-serving
    Why Boston's talent density and the Mass AI Coalition make it a real place to build deep tech
 
Chapters
00:00  Intro: the mattress company and a $300K run rate
02:00  OpenScout, OnePager, and the power law of fundraising
03:20  Open weights vs closed, and American AI leadership
05:17  Will open models catch the frontier?
07:03  Cybersecurity, jailbreaks, and the Hugging Face incident
09:41  Cephable and the push to on-device AI
10:21  Edge models: why reasoning compresses well
13:02  Career arc: IBM blockchain, Google Search, the founder bug
16:00  The Subconscious origin story (and a Guinness at Shy Bird)
17:50  Building inference for agents, not chatbots
20:12  Why the inference runtime is the right layer to own
22:51  Inside the tech: context compression and caching
25:42  Tim Run and the Tim models
26:15  The path to billions of agents
28:51  Why Boston: talent density and the Mass AI Coalition
31:45  Biggest lessons: nothing goes according to plan
33:30  Ruthless posting and the war on AI slop
36:29  Message to builders in the agent era
37:18  Rapid fire
39:46  The stat that stopped the room: 95% compression
40:52  Boston's under-the-radar company: Alpaca
42:15  The Avengers scenario: Suno and the Wei song
43:15  How Boston can help Subconscious
44:37  One big takeaway: agent abundance
 
GUESTS AND HOSTS:
 
Guest: Jack O'Brien, co-founder & CEO, Subconscious. https://www.subconscious.dev/
https://www.linkedin.com/in/thejackobrien/
Hosts: 
Matt Crane, MGMT Boston https://mgmtboston.com/
https://www.linkedin.com/in/matthewhcrane/
 
Will Lehmann, Step Function Investments. https://stepfunction.vc/
https://www.linkedin.com/in/wlehmann/
 
Watch/listen/Subscribe: 
 
Spotify: https://open.spotify.com/show/033HueFNlE1LPnEFPvVscs?si=e2351085b40b4334
Apple Podcasts: https://podcasts.apple.com/us/podcast/code-and-cobblestones/id6785977680
 
Produced in partnership with OBEY Creative, building podcast-driven authority content engines for B2B founders. Learn more at https://obeycreative.com

Andrew Lau grew up in the Bay Area, cut his teeth as an engineer at IBM and Microsoft, spent a decade at Endeca helping it grow from 7 people to 550 before Oracle bought it for over a billion dollars, and later ran strategy at Nanigans. He built Jellyfish in Boston anyway. 
In episode 5 of Code and Cobblestones, Andrew joins hosts Matt Crane and Will Lehmann to talk about what happens when half your pull requests get opened and closed by an AI agent, and why he thinks the software development lifecycle as we knew it is already gone.
The conversation runs from a stat that should worry every engineering leader (top companies went from 7% to nearly 50% of PRs handled by Claude in under a year) to what it even means for a model to be "Chinese" or "American" anymore, to the side project that quietly became Jellyfish's whole business.
What you'll learn:
Why Andrew says every engineer is now "a manager of agents, not humans," and what that does to team structure
How the ADLC (agentic development lifecycle) is replacing the SDLC, and what that shift actually looks like day to day
The real story behind Jellyfish's pivot, from a side project called Copilot Impact to the company's main business by the end of 2025
Why open weight models are everywhere in conversation but barely showing up in production
Andrew's bet on how AI model regulation actually plays out, and why brand trust matters more than model transparency
The honest tradeoffs of building in Boston instead of the Bay Area, including why he's still on a plane every three weeks
Lessons from mentors Steve Papa and Keith Johnson, and why customer development matters more now, not less
Timestamps:00:00 Andrew Lau Introduction
01:14 Building Real Relationships in Tech
02:05 Podcast Format and Hot Takes
02:34 Open Weight AI Models Explained
07:12 Why AI Cost Is Slowing Adoption
08:51 AI Policy and Data Sovereignty
11:17 Boston Startups and Physical AI
15:13 Mega Seed Rounds Explained
19:26 Andrew Lau Career Story
23:51 Why Jellyfish Was Founded
28:06 The Truth About AI Refounding
28:27 Copilot Impact Origin Story
30:17 From Side Project to Core Business
31:56 Every Engineer Is Now a Manager
34:43 Boston vs Silicon Valley
38:36 Startup Mentors and Focus
40:42 Why Customer Development Still Matters
42:55 AI Code Metrics Explained
45:38 Claude Writes 50% of Pull Requests
47:27 Final Thoughts on AI Transformation
 
Guest: Andrew Lau, Co-Founder and CEO of Jellyfish. Previously at Nanigans, and LoopIt. 
Resources:
Jellyfish Engineering Trends Report, July 2026
Jellyfish website: https://jellyfish.co/
Follow Andrew on LinkedIn: https://www.linkedin.com/in/amlau/
 
Hosts: Matt Crane, Founder of MGMT Boston and Will Lehmann Founder of Step Function
 
Code and Cobblestones is a love letter to Boston tech founders. New episodes spotlight the people building the city's next generation of companies.
 
▶ Subscribe for the rest of the tour. New episodes spotlight the founders and builders putting Boston on the map.
 
Produced in partnership with OBEY Creative, building podcast-driven content engines for B2B founders. One great conversation per week becomes 15-20 assets: clips, LinkedIn posts, blog content, and newsletter material. They close the Authority Gap: the distance between your expertise and the market's awareness of it.
Learn more at https://obeycreative.com/

Damien Lewke could build his company anywhere. He grew up in the Bay Area, threat-hunted on a nuclear weapon system at Northrop Grumman, helped define the EDR category at early CrowdStrike, closed nine-figure deals at Palo Alto Networks, and did his graduate research at MIT's CSAIL. He chose Boston. In episode 4 of Code and Cobblestones, Damien joins hosts Matt Crane and Will Lehmann to talk about founding Nebulock, a hunt-first security operations platform that just closed a $25 million Series A led by FirstMark, and why the next generation of cybersecurity gets built here.
 
The conversation runs from the open-source AI models rewriting the threat model, to the $40 he put in a bank account to start the company, to what a professional football career taught him about hiring.
 
What you'll learn
 
Why open-weight models like GLM 5.2 shift the advantage toward attackers, and what founders should do about it
What a "hunt-first" security platform actually does, explained plainly
The origin of Nebulock: 90 customers in 90 days, a 30-page manifesto, and a $40 bank account
How Damien hires, and the most underrated seed-stage hire right now
The real advantages of building a cybersecurity company in Boston
Three lessons from pro football that carry into founding a company
 
Chapters 
 
00:00 Episode 4, and our guest Damien Lewke 
01:46 Headlines: Mythos, Fable 5, and the open-source moment 
06:09 Two Boston raises: Nebulock and Scaled Cognition 
07:22 Inside the $25M Series A with FirstMark 
11:15 A World Cup detour: football lessons for founders 
15:32 From Northrop Grumman to early CrowdStrike 
20:02 Palo Alto, Australia, and back to MIT 
22:51 What a hunt-first security platform actually is 
24:06 The spark: a manifesto and a $40 bank account 
28:57 Three lessons: 90 customers, a high bar, get on a plane 
33:55 Why build in Boston? 
37:08 Rapid fire 
40:09 One message to cyber builders
 
Guest: Damien Lewke, Founder and CEO of Nebulock. Previously at Arctic Wolf, Palo Alto Networks, CrowdStrike, and Northrop Grumman. Graduate research at MIT CSAIL.
 
Hosts: Matt Crane, Founder of MGMT Boston and Will Lehmann Founer of Step Function VC
 
Mentioned: Nebulock, FirstMark, Step Function, CrowdStrike, Palo Alto Networks, Arctic Wolf, Northrop Grumman, MIT CSAIL, UCLA, Cribl, Bain Capital, Scaled Cognition, RSA, Veracode, Rapid7, Jellyfish, and Boston's security community.
 
Code and Cobblestones is a love letter to Boston tech founders. New episodes spotlight the people building the city's next generation of companies.
 
▶ Subscribe for the rest of the tour. New episodes spotlight the founders and builders putting Boston on the map.
 
Produced in partnership with OBEY Creative, a boutique authority media firm in the heart of Boston. We specialize in building Authority Capital for founder-led brands through podcasts and content engines that transform visionary builders into the primary voice of their industry.
Learn more at https://obeycreative.com/

Ellen Rubin has founded three companies, taken Netezza through its IPO, and sold startups to both Verizon and AWS. Now an operating partner at Glasswing Ventures, she sits down with Matt Crane and Will Lehmann to talk about why the hard, unglamorous parts of the stack are the ones worth building, and why Boston is the place to build them. From a citywide push to teach every high school student AI, to what a soccer crowd taught her about B2B sales, this one runs from the classroom to the cap table.
What you'll learn
Why Boston Public Schools' new AI literacy requirement matters, and how BostonPublic.ai (with Kayak's Paul English) is getting there first
How the "crossing the chasm" playbook broke, and what replaced it in the AI era
What makes a customer worth chasing, and how to measure real urgency instead of polite excitement
Why AI has to run through every part of a company, not just the product
The most underrated seed-stage hire right now
The real advantages of building in Boston (and the honest disadvantages)
What changes when you leave the operator seat for the other side of the table
Chapters 
00:00 A first guest, and why this one 00:56 Boston's new AI literacy requirement, and BostonPublic.ai 06:12 Three seed rounds: Enlay, Flux, and Pillar 10:01 What the Tartan Army taught Matt about B2B sales 14:34 The accidental entrepreneur: Booz Allen to Netezza 19:20 Founding CloudSwitch and ClearSky Data 23:07 Choosing your customer, and the chasm collapsing 28:43 The real advantages of building in Boston 31:27 Leaving the operator seat for Glasswing 34:18 Building an AI-native team in the age of agents 38:52 Rapid fire 42:50 Stop talking about what Boston doesn't have
Guest: Ellen Rubin, Operating Partner at Glasswing Ventures. Founder of CloudSwitch (acquired by Verizon) and ClearSky Data (acquired by AWS); early executive at Netezza through its IPO.
Hosts: Matt Crane and Will Lehmann.
Mentioned in this episode: Glasswing Ventures, BostonPublic.ai, Paul English (Kayak), the Mass AI Coalition, Enlay, Flux, Pillar, Allure Security, Realm Security, Whoop, and Boston Public Schools.
Code and Cobblestones is a love letter to Boston tech founders. New episodes spotlight the people building the city's next generation of companies.
▶ Subscribe for the rest of the tour. New episodes spotlight the founders and builders putting Boston on the map.
 
Produced in partnership with OBEY Creative, a boutique authority media firm in the heart of Boston. We specialize in building Authority Capital for founder-led brands through podcasts and content engines that transform visionary builders into the primary voice of their industry.
Learn more at https://obeycreative.com/