Jul 4, 2026 · 1h 11m · news
Open Models vs Frontier Models: Who Actually Wins? | The $100K Token Budget Every Engineer Will Need · 20VC with Harry Stebbings
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this engaging episode of 20VC, Sierra co-founder Clay Bavor shares strategic insights on AI model economics, building high-leverage enterprise software, and the unique operational philosophies that drive his multi-billion-dollar startup. He reflects on his 18-year career at Google, his close partnership with co-founder Brett Taylor, and how he balances intense company growth with a deep commitment to family and personal values.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 18.7% of the talking time here. How this is scored →
Harry as informed peer 3.6 Guest teaching 3.2 Guest disagreement 0.6 Harry pushing back 1.1
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 25:04 Rejection of 3.8% token spend benchmark
Clay explicitly rejects Harry's math on developer token spend being 3.8% of salary, stating that 3.8% is wildly off from where steady state will converge.
Hardest push from Harry ▶ 52:12 Challenging startup intensity with four kidsHarry directly challenges Clay's value of intensity by asking if it is literally possible to work as hard when supporting a family with four young children.
Biggest teaching moment ▶ 5:58 Explanation of foundation model capital expense realitiesClay educates Harry on the financial pitfalls of building foundation models, reframing them as perishable bags of floating point numbers that require ongoing capital expense.
Harry holds his own ▶ 24:25 Host calculates developer token spend percentages from Salesforce dataHarry uses Salesforce data ($300M Anthropic spend) and live calculations to demonstrate that token costs currently represent only 3.8% of developer salaries.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| The Alignment of the Planets to Start Sierra | 2 | 2 | 0 | 0 | Harry asks warm opening questions about Clay leaving Google after 18 years to launch Sierra with Brett Taylor. Clay cordially shares the alignment of timing and shared history that led to starting the company. | |
| Takeaways from 18 Years at Google | 3 | 4 | 1 | 1 | Harry asks about Google takeaways and probes whether Sierra considered pre-training foundation models. Clay educates on why training proprietary pre-trained models was a non-starter due to the huge capital expense of a perishable bag of floating point numbers. | |
| The Future of Open-Weights and Frontier AI Models | 4 | 4 | 1 | 3 | Harry questions whether open models advancing so fast weakens the case for expensive frontier models. Clay reframes by arguing that demand for top-tier frontier intelligence remains unbounded for complex tasks like legal and materials science. | |
| Token Economics, Reasoning Models, and GPU Constraints | 5 | 3 | 1 | 2 | Harry cites Nebius's founder regarding GPU compute supply constraints and challenges local model execution on mobile devices. Clay agrees on GPU constraints and explains thermal and compute limitations on phones. | |
| US vs. China: Scale Distillation and the Open Ecosystem | 5 | 4 | 1 | 2 | Harry cites Lovable reaching 500M ARR with 149 people to ask if enterprise companies like Sierra will need smaller teams. Clay points out that serving large Fortune 50 enterprises requires deep implementation and relationship management that lean consumer dev tools do not. | |
| Internal AI Tools: Pinecone and Sierra Brain | 3 | 3 | 0 | 0 | Harry asks detailed questions about Sierra's internal Pinecone agent and Sierra Brain. Clay explains their single MCP gateway and custom harnesses that streamline company operations. | |
| The Rise of the $100K Token Budget for Engineers | 6 | 4 | 2 | 3 | Harry calculates Salesforce's $300M Anthropic spend relative to developer salaries to challenge token spend projections. Clay forcefully rejects the 3.8% figure, asserting that token budgets will converge closer to 20% of developer spend. | |
| Retaining Product Focus in the Enterprise AI Market | 4 | 3 | 1 | 2 | Harry asks if being an enterprise business creates distance from product and asks about market maturation dynamics. Clay reframes the false choice by describing how he and Brett actively write code and build agents. | |
| Forward Deployed Engineering and Rapid Enterprise Deployment | 4 | 4 | 1 | 2 | Harry cites a previous guest claiming enterprise AI requires forward-deployed engineering. Clay explains how Sierra adopted Palantir's forward deployed model to compress deployment times to 6-8 weeks for giant enterprises. | |
| Scaling Beyond Support: Global Expansion and Lifecycle Management | 4 | 3 | 1 | 2 | Harry asks how Sierra expands beyond customer support into broader customer lifecycle management. Clay explains how customers like Rocket Mortgage use Sierra for inbound and outbound sales. | |
| Running Board Meetings on the Fast-Paced 'AI Time Clock' | 4 | 3 | 0 | 0 | Harry brings up Sierra running board meetings every six weeks instead of quarterly. Clay shares the rationale of operating on the fast-paced AI time clock and using written board memos over decks. | |
| Fundraising, Valuation, and Sierra's Core Values | 4 | 3 | 0 | 1 | Harry asks how Clay and Brett negotiate valuation when investors offer extreme prices. Clay explains prioritizing milestone capital over maximum valuation and intentionally taking lower prices than offered. | |
| Navigating 'Founder Mode' and Strategic Detail Engagement | 4 | 3 | 1 | 1 | Harry asks how founders navigate Founder Mode and decide where to micromanage. Clay defines selective engagement based on applying direct force where it moves the needle. | |
| Prioritizing Family While Maintaining Startup Intensity | 3 | 3 | 1 | 2 | Harry presses on whether a founder with 4 kids can literally work as hard as single founders. Clay describes extreme focus and his custom Starlink mini mobile setup for working during commutes. | |
| The Compounding Value of In-Person Work and Early-Career Learning | 3 | 4 | 0 | 0 | Harry asks about Clay's strong stance on in-person work. Clay explains the necessity of in-person mentorship for early-career professionals and cites Richard Hamming's compounding interest of knowledge. | |
| AI Mastery: Advice and AI-Native Hiring for Graduates | 3 | 3 | 0 | 0 | Harry asks advice for young graduates entering an AI workforce. Clay outlines Sierra's practical hiring interview where engineering candidates receive a $150 token budget to build using AI agents. | |
| AI and the Future of Cybersecurity | 4 | 3 | 1 | 1 | Harry frames a question around AI code generation sparking a golden age for cybersecurity. Clay addresses offensive vs defensive AI capabilities and touches on co-founder truth-seeking dynamics. | |
| The "Majors and Minors" Co-founder Dynamic | 3 | 3 | 0 | 0 | Harry asks who holds primary expertise on key topics between Clay and Brett. Clay breaks down their major and minor framework across sales, engineering, product, and operations. | |
| Underestimated Strengths of Google | 3 | 3 | 0 | 0 | Harry leads a quickfire round asking about lessons from Sundar Pichai and book recommendations. Clay highlights Sundar's dynamic range and recommends David McCullough's The Wright Brothers. | |
| Parenting Advice and Making Kids' Interests Your Own | 2 | 3 | 0 | 0 | Harry asks for parenting advice given Clay's four children. Clay highlights making your children's interests your own, describing how he learned basketball strategy for his eldest son. | |
| The Power of Partnership in Marriage | 2 | 2 | 0 | 0 | Harry asks about partnership in marriage and the kindest act Clay has experienced. Clay reflects on shared goals in marriage and his parents taking him to Macworld as a child. |
Statements from this episode (40)
Bavor: Unbounded demand for frontier AI remains broadly underappreciated
“We have not yet appreciated the unbounded demand for, call it, frontier levels of intelligence.”
Bavor: Chinese AI progress driven by willingness to distill frontier models
“Part of the driver of the difference is probably the willingness of Chinese companies to do scale distillation of the frontier models.”
Sierra deliberately took lower valuations than offered in every funding round
“Every one of our rounds, we actually guided to and took a lower price than we could have.”
Bavor: Sierra's most effective employees are 22- to 23-year-olds
“Some of our most effective employees of the entire company are 22 or 23 years old and have been completely AI-pilled.”
Sierra hired creator of foundational ReAct agent paper as head of research
“Our first founding head of research was the Princeton professor who literally wrote the paper on language model based agents, the react paper.”
Sierra will not pre-train foundation models, leaving capex to major AI labs
“And we're not, you know, we're not doing our own pre-training. We'll leave the capital expense there to, you know, the labs and the larger companies.”
Bavor: Pre-training foundation models is economically unviable for most startups
“The capital expense the ongoing capital expense to create what is effectively a highly perishable bag of floating point numbers just doesn't work. Just doesn't work for any, but a small number of companies.”
Sierra builds proprietary fine-tuned models on top of open weights
“So today we have a set of our own proprietary fine tune models. But these are fine tunes on top of open weights models.”
Bavor: Enterprises will mix open-weights and frontier models based on tasks
“And I think you'll end up with companies using both mixing and matching them depending on the task at hand.”
Bavor: Percentage of fully automated enterprise tasks is currently a rounding error
“Well, I think if you look at what percent of enterprise tasks are completely automated today, it's a rounding error, right? It's very low.”
Bavor: OpenAI's o1 model is an underrated AI breakthrough
“And I think actually one of the most underrated developments of the past few years was the O-one model from OpenAI in late 2024, where if you recall, there was a chart that showed, okay, test time, compute, or in amount of inference done, amount of thinking ou…”
Bavor: Local AI compute cannot solve server-side hardware bottlenecks
“Oh, it certainly wouldn't alleviate it. I think it will make some consumer applications much better. But, I mean, the reality is you need, you know, petaflops, exaflops of compute, certainly for training, and you want a whole bunch of compute quickly. At infer…”
Bavor predicts dedicated language model hardware in future phones and PCs
“Will you have a language model optimized hardware rolling out in, in our phones, in our computers? Yes.”
Bavor: US AI labs avoid open-weights models to protect frontier pricing
“I think if you have the US based labs and hyperscalers developing the frontier models. There's an obvious, you know, like, are they going to compete with themselves and drive, you know, price pressure on the frontier models by, you know, developing and releasi…”
Sierra engineers report 3x to 20x productivity boost from AI agents
“We have software engineers who are completely AI-pilled and using Cloud Code, Codex, our own internal agent we call Pinecone that we used to run much of the company on, and they estimate they are between three and 20 times more productive in terms of features …”
Bavor: 50% of Sierra's enterprise customers generate over $1B revenue
“We work with 40% of the fortune 50. We have 50% of our customers doing over a billion in revenue. We have 30% doing over ten billion in revenue.”
Bavor personally reviews and approves every single hire at Sierra
“So I, to date review and approve every single hire we make, and I get some help from pine cone and I've basically taught it.”
Bavor built an internal AI strategy partner trained on Sierra board letters
“Sierra brain is it starts with a 20 or 30 page document that grounds any agent in what we are as a company. What we do, how we're organized, our team structure, the competitive landscape, our strengths and weaknesses, all of these things. And then on top of th…”
Bavor: Top engineers spend over $100K annually on AI tokens
“I have heard and I have observed that top engineers who are really leaning in to cloud code codex and so on are spending more than a 100,000 dollars on a run rate basis on tokens per year.”
Bavor: CFOs will allocate dedicated per-employee AI token budgets
“So I think the direction that we're headed is some amount of Token budgeting on a per employee basis. I think for CFOs in the future, like capital allocation will look more like how do we allocate OpEx and not just OpEx and then headcount and headcount will be…”
Bavor: Software bottlenecks are moving from writing code to deciding what to build
“It used to be writing code. Now it's probably reviewing code. Pretty soon it will be deciding what is worth building and kind of editing kind of what could exist to what should exist.”
Bavor: AI token spend will reach 20% of software developer salaries
“I would not bet on 3.8%. I would bet on much closer to 20%. In software engineering, the gains, to me, seem unequivocally there.”
Sierra co-founder Bret Taylor still writes production code for the startup
“Brett is actually still an extraordinarily capable software engineer. It's remarkable. So, you know, some of the code that is in production, right, he has written”
Sierra expects to scale from hundreds of millions to billions of interactions
“We will in short order be in a way, one of the larger B to C companies. We're doing that via our customers, but you know, we'll, we'll, we're serving hundreds of millions of interactions, right? Soon billions of interactions.”
Bavor: Sierra is multiple times larger than its closest startup competitor
“We are at multiple larger multiple larger size than one of our next nearest similar vintage startup competitors growing faster.”
Bavor: Enterprise AI will be a duopoly with Sierra in the lead
“My hunch is it will be more like an Uber lift market. And we obviously think we're in the pole position to be the bigger of those two.”
Sierra built its initial platform embedded inside Weight Watchers, Sonos, and SiriusXM
“We then enlisted half a dozen design partners that we built the first version of our product and platform with and for. And these are, in the history of the company, legendary, legendary companies. Olokai, great flip-flops, you should buy them. SiriusXM, Sonos…”
Sierra aims for far greater customer volume than Palantir
“My understanding is, you know, Palantir has kind of low hundreds of customers. We are and intend to be at a lot larger scale than that.”
Bavor: Sierra holds board meetings on a six-week cadence
“You mentioned the six week cadence. We have kind of a tick tock, a three hour meeting and a one and a half hour meeting. We've done this since the beginning of the company”
Sierra uses six-to-ten-page written memos instead of decks for board meetings
“As for running the board meetings themselves, we don't have board decks, we have board memos. So Brett and I write a usually six to 10 page memo.”
Sierra founders personally monitored every AI customer conversation during Black Friday
“When we, in our first Black Friday, Cyber Monday with a set of retailers, one of one of our lead engineers, our head of operations, me or Brett, was in real time personally reading every single conversation that our agents were having, because we wanted to mak…”
Bavor: Conversational AI agents will inevitably replace websites as primary customer interfaces
“I think there is an inevitability to companies interacting with their customers via really sophisticated agents that capture all that they know and all they can do on behalf of their customers and get the job done on their behalf that handle the complexity as …”
Bavor uses dual cellular and Starlink Mini for uninterrupted commute connectivity
“I now have a very complex networking setup that combines two cellular networks and a Starlink mini so that I have uninterrupted beautiful connectivity to and from the city every day.”
Bavor calls Richard Hamming's 'You and Your Research' essential reading for graduates
“There's a talk that I love by the renowned computer scientist, Richard Hamming, you and your research. And it's for any new graduate, probably the single best thing on a per word basis, I think you can read.”
Sierra gives engineering candidates $150 in tokens for AI-native interviews
“We completely changed our engineering interview process. So it now looks much more like, here's a very here's a kind of prompt think through an application you would like to build. Cool. Okay. Here's a 150 dollars to spend on choose your coding agent. You can …”
Bavor: Every Sierra job interview will be AI-native within two months
“I will be disappointed if in the next no more than two months, not every one of our interviews has some strong AI native component to it.”
Clay Bavor details recent operational debate with Sierra co-founder Bret Taylor
“I was on one side. It was like, I think we need better kind of process and structure around this thing. Brett was on the side of people and maybe we need different leaders or a different leader in this space. The answer is with most things like turned out to b…”
Bavor: Co-founders should use a 'majors and minors' responsibility framework
“We think about rather than dividing up the company, we think about Majors and minors for every part of the company.”
Bavor: Sundar Pichai's dynamic range in strategic thinking is second to none
“Sundar has a remarkable ability to look at a problem From wildly different zoom levels. His dynamic range in thinking is second to none. Zoomed all the way out. Highest level strategy. How is this going to unfold over the next five years? All the way into the …”
Bavor: 'The Wright Brothers' is the most accurate portrait of entrepreneurship written
“David McCullough, The Wright Brothers. It's so good. It's so good. It's a tight history of obviously the invention of the first heavier than air aircraft. And to me, it is as accurate a portrait of entrepreneurship and invention as has been written anywhere.”