Tomasz Tunguz

Theory Ventures

Student of startups, author of tomtunguz.com, investor, & partner to founders.

General Partner

Redpoint Ventures

Privileged to support the journey of early stage SaaS, data and infrastructure founders. Worked with Looker ($2.6b M&A), MotherDuck, Mysten Labs (SUI), Monte Carlo, Omni, Chorus ($575m to ZI), Expensify (IPO), ThredUp (IPO), Kustomer ($1b sale to META), & others.

Managing Director

Winning with Data

I co-authored Winning with Data with Looker CEO Frank Bien.  Winning with Data: Transform Your Culture, Empower Your People, and Shape the Future explores  the cultural changes big data brings to business, and shows you how to adapt your organization to leverage data to maximum effect.

Co-Author

Google, Inc.

Managing AdSense for Social Networks including the $900M MySpace deal.
Managed AdSense International including product launches in Greece, Israel, Thailand, Indonesia and Bulgaria

Product Manager

Appian Corporation

Architected, developed and deployed systems for the US Government serving DHS

Engineer

Student of Startups
Backer of 9 unicorns
Author of http://tomtunguz.com 
Subscribe https://ttunguz.kit.com/ef0a779b9c

Founder

70% of production AI teams use open source models. 72.5% connect agents to databases, not chat interfaces. This is what 375 technical builders actually ship - & it looks nothing like Twitter AI.

70% of teams use open source models in some capacity. 48% describe their strategy as mostly open. 22% commit to only open. Just 11% stay purely proprietary.

Agents in the field are systems operators, not chat interfaces. We thought agents would mostly call APIs. Instead, 72.5% connect to databases. 61% to web search. 56% to memory systems & file systems. 47% to code interpreters.

The center of gravity is data & execution, not conversation. Sophisticated teams build MCPs to access their own internal systems (58%) & external APIs (54%).

Synthetic data powers evaluation more than training. 65% use synthetic data for eval generation versus 24% for fine-tuning. This points to a near-term surge in eval-data marketplaces, scenario libraries, & failure-mode corpora before synthetic training data scales up.

The timing reveals where the stack is heading. Teams need to verify correctness before they can scale production.

88% use automated methods for improving context. Yet it remains the #1 pain point in deploying AI products. This gap between tooling adoption & problem resolution points to a fundamental challenge.

The tools exist. The problem is harder than better retrieval or smarter chunking can solve.

Teams need systems that verify correctness before they can scale production. The tools exist. The problem is harder than better retrieval can solve.

Context remains the true challenge & the biggest opportunity for the next generation of AI infrastructure.

Explore the full interactive dataset here : https://lnkd.in/gZZuFHbp or read Lauren’s complete analysis : https://lnkd.in/gHBrdbCT.

Tomasz is a General Partner at Theory Ventures, a $235m early stage venture capital firm. He blogs sat tomtunguz.com & co-authored Winning with Data. 

He has worked or works with Looker, Kustomer, Monte Carlo, Dremio, Omni, Hex, Spot, Arbitrum, Sui & many others.

Venture Capitalist at Theory

I remember hosting a dinner of sales leaders to talk about AI. I asked them what will your CRM do for you in the future?

Nearly unanimously came the reply : enable salespeople to spend most of their time with customers.

Private equity firms have emerged as the newest distribution channel for AI startups.

While public companies have decreased from 6,639 in 2000 to 3,550 in 2024, PE-owned companies in the US have grown from 1,950 to 14,300. The rate of growth continues to accelerate.

The crossover happened in 2009, when PE inventory overtook public company counts for the first time. By 2024, PE-backed companies outnumber public firms by roughly 4:1.

The shift is driven by the massive expansion of PE ownership across corporate America.

That’s not to say the sizes of PE-owned companies are the same as publics. In fact, they are smaller.

The point isn’t that startups previously focused on public companies. Rather, the data reveals the immense scale of private equity portfolios.

(second image)

Data Sources : CRSP, Wilshire 5000, PitchBook, American Investment Council, Citizens Bank.

The mid-market profile of these PE-owned companies suits AI startups’ desires for faster sales cycles.

Plus, the profit motive of private equity aligns perfectly with AI startups’ capacity to cut costs & drive efficiency. PE firms acquire companies to improve margins & operational performance before exit.

AI tools that reduce headcount, automate processes, or accelerate workflows deliver exactly what PE operating partners need.

A private equity firm owning 25 companies proves value in one or two before rolling out to the entire portfolio. Control enables rapid deployment. This creates an efficient channel for AI startups to demonstrate value & cross-sell.

PE firms gain operational leverage while AI startups access more than 14,000 motivated buyers. This new go-to-market motion redefines how AI software reaches the market, bypassing the traditional enterprise sales grind in favor of networks that can deploy at scale.

Office Hours: GTM Leadership in the AI Era: Pricing, Hiring, and Compensation

Thursday, December 11
5:30–7:30 PM PDT

As AI-native companies scale, GTM leaders are being forced to rethink long-standing assumptions about pricing, hiring, incentives, and how revenue organizations should operate in a world defined by rapid iteration and usage-driven value. For this month’s Office Hours, we’re bringing together two leaders who have repeatedly built high-performing GTM functions at some of the fastest-growing companies in AI and infrastructure.

Joining us are David Morse and Graham Moreno of Cognition AI, operators with deep experience building category-defining revenue teams. Graham has led go-to-market organizations from pre-seed through hundreds of millions in ARR, including scaling Windsurf more than 40x in 18 months and helping build Grafana Labs from the earliest days to over 250M in ARR. David brings nearly a decade of experience deploying AI into large enterprises, having led Scale AI’s core business as VP/GM, guided Hebbia through its growth stage as CRO, and built Dropbox’s enterprise field teams through its IPO.

Their collective experience, spanning early hypergrowth, enterprise AI adoption, pricing innovation, and revenue operations, will ground a candid conversation on what GTM excellence looks like in the AI era.

This Office Hours session will explore:

Pricing AI
How AI-native companies are approaching pricing strategy, from usage-based and value-based models to new frameworks that capture ROI and set clear customer expectations as capabilities evolve.

Building the AI-Native GTM Organization
What it takes to identify and attract the next generation of GTM leaders who can navigate rapid product iteration, shifting pricing models, and AI-driven buyer behavior.

Rethinking Compensation and Quotas
How traditional sales compensation is being reinvented as automation, usage variability, and rapid scaling redefine performance metrics and incentives.

This session is designed for founders, GTM leaders, and operators building revenue organizations in the AI era. We hope you will join us for a candid, tactical conversation with two of the sharpest GTM minds in AI.

Since I watched software engineers using AI, I’ve become jealous.
I’ve seen the most sophisticated software engineers assign 15 to 20 coding tasks in GitHub to an AI. They play foosball. They grab coffee. They return to evaluate the agent’s work.

While OpenAI signed $1.15 trillion in compute contracts through 2035, DeepSeek trained a frontier model for $6 million. This was 2025’s central question : are we building on bedrock or quicksand?

Two revelations this week have shaken the narrative in AI : Nvidia’s earnings & this tweet about Gemini.

The AI industry spent 2025 convinced that pre-training scaling laws had hit a wall. Models weren’t improving just from adding more compute during training.

Enterprises learned a lesson from cloud data warehouses. They handed over both data & compute, then watched as the most strategic asset in their business, how they operate, became someone else’s leverage, which created an opportunity for Iceberg.

70% of production AI teams use open source models. 72.5% connect agents to databases, not chat interfaces. This is what 375 technical builders actually ship — & it looks nothing like Twitter AI.

Enterprises learned a lesson from cloud data warehouses. They handed over both data & compute, then watched as the most strategic asset in their business, how they operate, became someone else’s leverage, which created an opportunity for Iceberg.

Fool me once…

Leaders have recognized their companies need a new system of record for AI agents in the form of a context database. There are two different kinds of these context databases : (first image)

Operational context databases store standard operating procedures & institutional knowledge : when a customer calls about resetting a password, when legal reviews an NDA with a new prospect, when HR answers questions about options vesting for a new hire.

All of these processes represent trade secrets & intellectual property, which are key assets for a business. Capturing them from employees ensures continuity in processes & builds a sustainable, defensible asset.

Analytical context databases are a semantic evolution of semantic layers : they contain definitions & calculations for metrics like revenue or customer acquisition cost.

Semantic layers told AI what data meant. Analytical context databases teach AI how to reason about it.

Steven Talbot’s recent piece on Omni’s agentic analytics architecture describes :

a coordinator mechanism, which decides which tool to use next based on the question, the results, & what’s already been tried.

The key to both operational & analytical context databases isn’t the databases themselves. It’s the feedback loops within them.

Steven’s system adapts mid-flight, retries when things break, or stops when it has something useful to show. This creates an ever-improving cycle of accuracy. Accuracy creates trust. Trust creates adoption. Adoption creates more feedback. Companies that develop the best feedback loops will build the most valuable context databases.

Context databases enable the future of process automation, representing the real promise of AI within the workforce. It’s the evolution of RPA (robotic process automation), but it’s RPA & process discovery injected with non-determinism.

This non-determinism is essential for the success of AI agents. It allows for exception handling, forestalling one of the failure modes of the first generation of RPA. AI agents are excellent at ingesting large volumes of content & reasoning about them.

The move from manual context engineering to automated context platforms is inevitable. Context databases will be sold as standalone products & bundled. Enterprises will come out of this transformation for the better : with evolving systems that improve over time.

Private equity firms have emerged as the newest distribution channel for AI startups.

The AI market today is bacon in a hot skillet. Everything is sizzling, moving, & changing at an incredible pace. We’re all watching it closely.

Market share is fluid because no one yet knows what AI can do & the second we think have grasped it, models improve. The Nvidia chip performance & the launch of Gemini 3the biggest gain ever in Google model performance suggest no simmering ahead.

As long as the underlying models hurtle towards PhD level performance, people will continue to test. How much better is Gemini 3 at coding? tool calling? writing?

If the progress is material, then the benefit of switching is worth the activation energy.

Today, startups, incumbent software companies, cloud providers & AI labs all are competing. First the model, then infrastructure (memory & retrieval), then tools, then applications. Will the foundational models play at the application layer? Or will the applications differentiate themselves enough to overcome model differences?

Who can take advantage of the next big leap in model performance fastest? Which sales team can reach the target customers first & write the RFP?

This is the Great Game of Risk in Category Creation & aggression wins.

But this era of fluidity won’t last forever. The rate of improvement in AI models will eventually attenuate. When the performance gap between the best model & the second-best model shrinks, the incentive to switch evaporates.

Switching costs will start to matter more than marginal performance gains. The custom tools I’ve built, the muscle memory I’ve developed, the integrations my company has deployed, the enterprise contracts signed, all inertia.

At that point, the fat begins to congeal.

The winners will be those who use the sizzling phase to build fat worth congealing around.

This fun analogy came up during my conversation with Harry, Jason, & Rory.

https://lnkd.in/gixvTmGe

The AI market today is bacon in a hot skillet. Everything is sizzling, moving, & changing at an incredible pace. We’re all watching it closely.

Market share is fluid because no one yet knows what AI can do & the second we think have grasped it, models improve. The Nvidia chip performance & the launch of Gemini 3the biggest gain ever in Google model performance suggest no simmering ahead.

I remember hosting a dinner of sales leaders to talk about AI. I asked them what will your CRM do for you in the future?

Nearly unanimously came the reply : enable salespeople to spend most of their time with customers.

Listening to Lenny’s Podcast with Jeanne Grosser, COO at Vercel, I discovered how some companies are achieving that milestone by transforming their inbound sales teams in six weeks.

"For us, we had 10 SDRs doing this inbound workflow, & now we just have one that is effectively QAing the agent. The other nine, we deployed on outbound."

"It was six weeks before we felt confident going from 10 to 1. So it wasn’t like this was a multi-quarter process. It actually moved super quickly."

"We were able to hold that lead to opportunity conversion rate flat. So the agent is as good as our humans were."

In retrospect, it should be obvious that speed creates an unexpected advantage. A sales representative, or an AI sales representative, can respond & analyze a lead at any point during the day. It’s important to meet a customer at the point of maximum interest to capitalize on their inertia or excitement.

"t’s actually condensed the number of touches it takes to convert because it’s so much quicker at responding relative to leads inevitably sitting in the queue or coming in at nighttime & no one can get to it."

The build? One engineer, part-time.

"The person who built the lead agent was a single GTM engineer. He spent maybe 25, 30% of his time on this."

The target that every sales leader at that dinner articulated is now within reach.

"I think we’re getting to a point where with layering in agents, ideally, we finally get salespeople to a point where they’re actually spending 70% of their time interacting with humans."

Thirty percent customer-facing time becomes seventy percent. Double the human contact means salespeople developing deeper relationships, driving more success for customers, & less on administration.

That’s the promise of AI in the workplace.

Streaming is the next category to consolidate within the modern data stack. IBM announced its intent to acquire Confluent.

Streaming is the next category to consolidate within the modern data stack. IBM announced its intent to acquire Confluent.

The deal values Confluent at $11.1 billion, or 10.0x LTM revenue. Confluent commands more than 40% of the Fortune 500 as customers & has grown into a $1.1 billion revenue business.

The founders of Confluent, including CEO Jay Kreps, created Apache Kafka, a streaming technology built inside LinkedIn. Founded in 2014, Apache Kafka now runs at more than 80% of the Fortune 100. Kafka powers real-time data pipelines & stream processing, updating data systems whenever a new event happens. When a taxi ride is booked, a credit card is swiped, or a user likes a comment, Kafka handles the data flow.

Confluent continues to grow nicely, with recent quarterly revenue of $298.5 million, growing 19.3% year over year. It has a gross margin of 74.1%, which is typical for software companies. Although its operating margins are negative, roughly -27%, this reflects a high cost of sales.

The company’s sales efficiency sits at 0.38x : Q3’s marginal gross profit of $13.5M annualized ($54M) divided by Q2’s selling & marketing expenses of $143.6M. For every dollar spent on S&M in one quarter, the company generates 38 cents in incremental gross profit the next year.

New customer acquisition reinforces the challenge.

Confluent added 36 net new $20K+ customers in Q3. The $100K+ cohort represents the largest sequential increase in 2 years.

Growth comes primarily from expansion : net revenue retention sits at 114%, meaning existing customers increase spend by 14% annually. Gross retention hovers near 90%. The $100K+ customers account for more than 90% of ARR. Confluent’s challenge isn’t keeping customers. It’s acquiring new ones.

The most compelling trend : 10-percentage-point YoY improvement in operating margins, suggesting a path to profitability within 12-18 months.

Is Confluent a bargain or appropriately priced?

The bull case points to category-defining technology in Apache Kafka, 21%+ growth with improving margins, a clear path to profitability & critical infrastructure for AI & real-time applications. The bear case notes the company remains unprofitable, faces competitive threats from AWS Kinesis & Azure Event Hubs, competes with open-source Kafka alternatives & carries IBM integration execution risk.

A major question for Confluent over the last few quarters has been : what is the next adjacent product category to catapult another wave of growth? The AI surge has been a positive for the company.

Within IBM, Confluent may find its technology complementary to the raft of AI technologies demanded by large enterprises.

No, founders are not adopting Bryan Johnson’s regimen to reverse aging. Quite the opposite : the average founder raising capital ages six months every year.

Two revelations this week have shaken the narrative in AI : Nvidia’s earnings & this tweet about Gemini.

The AI industry spent 2025 convinced that pre-training scaling laws had hit a wall. Models weren’t improving just from adding more compute during training.

Then Gemini 3 launched. The model has the same parameter count as Gemini 2.5, one trillion parameters, yet achieved massive performance improvements. It’s the first model to break 1500 Elo on LMArena & beat GPT-5.1 on 19 of 20 benchmarks.

Oriol Vinyals, VP of Research at Google DeepMind, credited improving pre-training & post-training for the gains. He continued that the delta between 2.5 & 3.0 is as big as Google has ever seen with no walls in sight.

This is the strongest evidence since o1 that pre-training scaling still works when algorithmic improvements meet better compute.

Second, Nvidia’s earnings call reinforced the demand.

"We currently have visibility to $0.5 trillion in Blackwell and Rubin revenue from the start of this year through the end of calendar year 2026. By executing our annual product cadence and extending our performance leadership through full stack design, we believe NVIDIA will be the superior choice for the $3 trillion to $4 trillion in annual AI infrastructure build we estimate by the end of the decade."

"The clouds are sold out and our GPU installed base, both new and previous generations, including Blackwell, Hopper and Ampere is fully utilized. Record Q3 data center revenue of $51 billion increased 66% year-over-year, a significant feat at our scale."

The infrastructure is accelerating headlong into hundreds of billions next year & Nvidia predicts it will be in the trillions, citing “$3 trillion to $4 trillion in data center by 2030”.

As Gavin Baker points out, Nvidia confirmed Blackwell Ultra delivers 5x faster training times than Hopper.

Gemini 3 proves the scaling laws are intact, so Blackwell’s extra power will translate directly into better model capabilities, not just cost efficiency.

Together, these two data points dismantle the scaling wall thesis.

A new study from OpenAI shows AI saves the average white-collar worker 54 minutes per day. Where does all that value go?

BLS data shows median weekly earnings for full-time workers hit $1,165 in Q3 2024, or $60,580 annually. Across 2,080 working hours, hourly compensation equals $29.13. One hour saved per day equals 250 hours per year, or $7,282 in recovered productivity per seat.

Current AI pricing captures between 3% & 5% of this value : (second image)

Typically, vendors capture 10-15% of value, leaving employers & employees with 85-90% value capture.

This gap suggests significant pricing power remains untapped since the total value capture for these tools is only 5%. Microsoft announced price increases last week, with Microsoft 365 E3 rising from $36 to $39 per user monthly starting July 2026 - an 8.3% increase, closer to a COLA adjustment than a value-based price increase.

Bundling complicates the picture.

Microsoft 365 Copilot & Google Workspace AI include access to email, spreadsheets, & presentation software. ChatGPT Plus does not. Will enterprises pay for standalone AI applications on top of their existing Copilot or Workspace subscriptions? Must standalone AI tools also bundle these features?

Unbundling is already happening in some categories too :

Gamma, a $2.1B unicorn building AI presentation software, charges $480 per seat annually. 33% more than Microsoft 365 Copilot, which already includes PowerPoint with AI features.

Gamma has 600,000+ paying subscribers, suggesting a market exists for best-in-class vertical tools even when bundled alternatives exist.

In the SaaS ecosystem, bundling & unbundling were both motifs. AI doesn’t seem any different, aside from the significant productivity gains that create room for both strategies to win.

A new study from OpenAI shows AI saves the average white-collar worker 54 minutes per day. Where does all that value go?

taplio

Tomasz Tunguz's Linkedin Analytics

Tomasz Tunguz

Performance ScoreTomasz Tunguz's LinkedIn Performance

What is Tomasz talking about?

Who is engaging with Tomasz

Tomasz Tunguz's Best Posts (last 30 days)

Ready to grow your LinkedIn presence?

taplio

.css-6ytdur{background-image:linear-gradient(284deg, rgba(39, 43, 72, 1) 0%, rgba(33, 47, 183, 1) 100%);color:transparent;-webkit-background-clip:text;background-clip:text;}Tomasz Tunguz's Linkedin Analytics

Tomasz Tunguz

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Tomasz Tunguz's LinkedIn Performance

What is Tomasz talking about?

Who is engaging with Tomasz

Tomasz Tunguz's Best Posts .css-1emoo84{font-size:var(--chakra-fontSizes-md);font-weight:400;color:var(--chakra-colors-gray-600);}(last 30 days)

Ready to grow your LinkedIn presence?

Tomasz Tunguz's Linkedin Analytics

Performance ScoreTomasz Tunguz's LinkedIn Performance

Tomasz Tunguz's Best Posts (last 30 days)