Barrchives

Temporal’s co-founders join Barr Yaron and Lenny Pruss to unpack how durable execution became the backbone of modern distributed apps and why it’s a perfect fit for AI agents. Samar and Max trace the path from Amazon SWF to Uber’s Cadence to founding Temporal, dig into developer experience choices, hard lessons with Cassandra, and what “code that can’t crash” really means in practice. This episode also covers open source strategy, multi-agent orchestration, Nexus RPC, how startups and enterprises are adopting Temporal, and what scaling the company taught them as leaders.

If you ship backend systems, build AI agents, or care about reliability at scale, this one’s for you.

This episode is broken down into the following chapters:

• 00:00–00:06 — Origins: Samar and Max, Barr and Lenny, why this convo

• 00:06–00:14 — Early systems: Amazon SWF → Azure Durable Task; DX lessons

• 00:14–00:23 — Uber years: replacing Kafka, “Jeremy,” birthing Cadence, open-sourcing from day one

• 00:23–00:31 — Durable execution, explained; code-first over DSLs; SDK ergonomics (Go/Java)

• 00:31–00:36 — Hard tech war stories: Cassandra, queues on Cassandra, multi-region replication

• 00:36–00:45 — AI agents ≈ dynamic workflows; why Temporal fits agents and tools

• 00:45–00:59 — Roadmap: streaming, large payloads, data workflows, Nexus RPC for long-running calls

• 00:59–01:13 — Adoption & GTM: digital natives, fintech, startups; greenfield AI vs brownfield; “Temporal is overkill?”

• 01:13–01:22 — Case study: OpenAI Codex on Temporal; internal dogfooding

• 01:22–01:40 — Scaling the org vs. scaling the tech; remote shift, hiring; mission & next five years

Subscribe to the Barrchives newsletter: https://www.barrchives.com/

Spotify: https://open.spotify.com/show/37O8Pb0LgqpqTXo2GZiPXf

Apple: https://podcasts.apple.com/us/podcast/barrchives/id1774292613

Twitter: https://x.com/barrnanas

LinkedIn: https://www.linkedin.com/in/barryaron/

Factory co-founder and CEO Matan Grinberg joins Barr Yaron to talk about the future of agent-driven development, why enterprise migrations are the perfect wedge for AI adoption, and how software engineering is moving toward a world where humans orchestrate instead of implement.

They dive into Factory’s origin story, the challenges of building AI systems for large organizations, and what the world might look like when millions of “droids” (AI agents) collaborate on software. Along the way, Matan shares surprising use cases, lessons from working with enterprises, and how his personal journey—from physics to burritos to building Factory—has shaped his leadership.

This episode is broken down into the following chapters:

00:00 – Intro and welcome

01:06 – Founding Factory: from ChatGPT experiments to AI engineers in every tab

04:05 – Early vision: autonomy for software engineering

06:14 – Why focus on the enterprise vs. indie developers

08:29 – Behavior change and technical challenges in large orgs

10:25 – Using painful migrations as a wedge for adoption

12:20 – The paradigm shift to agent-driven development

15:59 – Ubiquity: making droids available across IDEs, Slack, Jira, and more

17:16 – Why droids need the same context as human engineers

20:15 – Memory, configurability, and organizational learning

23:05 – How many droids? Specialization vs. general purpose agents

25:34 – Bespoke vs. common workflows across enterprises

27:06 – The hardest droid to build: coding itself

28:26 – Testing, costs, and scaling agentic workflows

30:29 – Why observability is essential for trustworthy agents

31:28 – Surprising use cases: PM adoption and GDPR audits

34:02 – Who Factory is building for: PMs, juniors, seniors, and beyond

36:09 – Systems thinking as the core engineering skill

38:09 – Building for enterprise trust: guardrails and governance

40:35 – What’s missing at the model layer today

42:43 – Migrations as a go-to wedge in go-to-market

43:53 – The thought experiment: what if 1M engineers collaborated?

46:07 – Scaling agent orgs: structure, monitoring, and observability

48:46 – Why everything must be recorded for droids to succeed

50:11 – Recruiting people obsessed with software development

51:37 – Burritos, routines, and how Matan has changed as a leader

53:41 – From coffee to Celsius, and why team culture matters most

54:20 – Closing thoughts: the future when agents are truly ubiquitous

Subscribe to the Barrchives newsletter:

https://www.barrchives.com/

Spotify:

https://open.spotify.com/show/37O8Pb0LgqpqTXo2GZiPXf

Apple: https://podcasts.apple.com/us/podcast/barrchives/id1774292613

Twitter: https://x.com/barrnanas

LinkedIn: https://www.linkedin.com/in/barryaron/

Datadog CEO Olivier Pomel joins Barr Yaron and Sunil Dhaliwal to discuss the evolution of observability, the role of AI inside Datadog, and how the future of software development will be shaped by agents, voice interfaces, and new approaches to monitoring.

From Datadog’s origins in the cloud shift to its latest AI-driven products like Bits AI, Olivier shares how the company is building for the future while staying deeply connected to its customers’ real problems.

Spotify:

https://open.spotify.com/show/37O8Pb0LgqpqTXo2GZiPXf

Apple: https://podcasts.apple.com/us/podcast/barrchives/id1774292613

Twitter: https://x.com/barrnanas

LinkedIn: https://www.linkedin.com/in/barryaron/

In this episode of Barrchives, Barr Yaron sits down with Simon Eskildsen, co-founder and CEO of turbopuffer, to explore how he went from infrastructure challenges at Shopify to launching a groundbreaking vector database company.

Simon shares his journey from recognizing the inefficiencies of traditional vector storage solutions to creating TurboPuffer, a revolutionary database designed specifically for AI-driven applications. He details key moments of insight—from working with startups struggling with prohibitive storage costs, to realizing the untapped potential of affordable object storage combined with modern vector indexing techniques.

This episode is broken down into the following chapters:

• 00:00 – Intro: Simon Eskildsen, Founder of TurboPuffer

• 00:26 – The “aha” moment: Simon’s transition from Shopify and startup consulting to founding TurboPuffer

• 03:13 – Turning “strings into things”: The power of vector search

• 05:51 – Why vector databases? Economic drivers and technology shifts

• 07:35 – Building TurboPuffer V1: Key architecture choices and early trade-offs

• 10:44 – Challenges of indexing: Evaluating exhaustive search, HNSW, and clustering

• 17:23 – Finding product-market fit with Cursor: TurboPuffer’s first major customer

• 20:05 – Defining TurboPuffer’s ideal customer profile and market positioning

• 23:43 – Gaining conviction: When Simon knew TurboPuffer would scale

• 25:39 – TurboPuffer V2: Architectural evolution and incremental indexing improvements

• 32:12 – How AI-native workloads fundamentally change database design

• 35:41 – Key trade-offs in TurboPuffer’s database architecture (accuracy, latency, and cost)

• 38:07 – Ensuring vector database accuracy: Production vs. academic benchmarks

• 41:03 – Deciding when TurboPuffer was ready for General Availability (GA)

• 42:27 – The future of vector search and storage needs for AI agents

• 45:03 – Building customer-centric engineering teams at TurboPuffer

• 47:12 – Common storage hygiene mistakes (or opportunities) in AI companies

• 49:42 – Simon’s personal growth as a leader since founding TurboPuffer

Subscribe to the Barrchives newsletter: https://www.barrchives.com/

Spotify: https://open.spotify.com/show/37O8Pb0LgqpqTXo2GZiPXf

Apple: https://podcasts.apple.com/us/podcast/barrchives/id1774292613

Twitter: https://x.com/barrnanas

LinkedIn: https://www.linkedin.com/in/barryaron/

AI won’t fix healthcare unless it starts with the conversation.

In this episode, Zachary Lipton—Chief Technology & Science Officer at Abridge and Raj Reddy Associate Professor of Machine Learning at Carnegie Mellon University—joins Barr Yaron for a deep, technical, and emotional dive into how AI can truly transform clinical care.

From building a world-class ambient documentation system to tackling speech recognition in 28 languages, Zack shares what it takes to engineer trust into AI when the stakes are patient lives, not just clicks.

We cover:

• Why general-purpose models fail in clinical settings

• How Abridge designs for accuracy, context, and trust

• The tension between personalization and evaluation

• Why ambient AI might be the most promising foundation for fixing healthcare

This is one of the most in-depth looks at what it actually takes to build production-grade AI in medicine.

This episode is broken down into the following chapters:

00:00 – Intro

00:34 – What Abridge actually does (hint: it’s not just notes)

01:09 – Why documentation is killing the healthcare experience

03:05 – How we got to the current burnout crisis

04:16 – The key insight: healthcare is a conversation

07:33 – Building a digital scribe: the original vision for Abridge

09:15 – Why off-the-shelf models don’t cut it in clinical speech

11:36 – 28 languages, noisy ERs, and overlapping conversations

13:20 – Predicting what enters the medical lexicon next

14:21 – How Abridge adapts models for edge-case medical speech

15:18 – Beyond transcripts: the complexity of clinical note generation

17:10 – Foundation models are tools, not solutions

18:06 – The “Ship of Theseus” strategy of model orchestration

20:32 – Style transfer for doctors, patients, and payers

20:54 – Metrics: ASR evaluation vs. documentation quality

23:43 – Stratifying ASR performance by setting, language, and jargon

24:50 – Why eval is so hard when there’s no “gold note”

25:45 – The tension between personalization and general eval

28:05 – Lessons from machine translation: building robust eval pipelines

30:32 – Abridge’s “look at the f*cking data” (LFD) internal review

33:54 – Blinded clinical eval with linked evidence and audio

36:50 – Why human fallibility is just as real as AI hallucination

38:21 – What kind of CTO Zack actually is

40:32 – Why AI product development is its own discipline

42:44 – AI innovation now lies in the product-data-model loop

44:25 – Closing the loop: how design drives modeling

45:25 – How Abridge hires researchers who care about product

47:29 – The mission filter: if you’d be equally happy at Microsoft, go

49:35 – What’s next: the AI layer for healthcare, not point solutions

52:57 – Closing thoughts

Subscribe to the Barrchives newsletter: https://www.barrchives.com/

Spotify: https://open.spotify.com/show/37O8Pb0LgqpqTXo2GZiPXf

Apple: https://podcasts.apple.com/us/podcast/barrchives/id1774292613

Twitter: https://x.com/barrnanas

LinkedIn: https://www.linkedin.com/in/barryaron/

What if deep learning isn’t the future of AI—but just part of it?

In this episode, Ori Goshen, Co-founder and Co-CEO at AI21 Labs, shares why his team set out to build reliable, deterministic AI systems—long before ChatGPT made language models mainstream.

We explore the launch of Wordtune, the development of Jamba, and the release of Maestro—AI21’s orchestration engine for enterprise agentic workflows. Ori opens up about what it takes to move beyond probabilistic systems, build trust with global enterprises, and balance research and product in one of the most competitive AI markets in the world.

If you want a masterclass in enterprise AI, model training, architecture tradeoffs, and scaling innovation out of Israel—this is it.

🔔 Subscribe for deep dives with the people shaping the future of AI.

This episode is broken down into the following chapters:

00:00 – Intro

00:47 – Why AI21 started with “deep learning is necessary but not sufficient”

02:34 – Building reliable AI systems from day one

03:46 – The risk of neural-symbolic hybrids and early bets on NLP

05:40 – Why Wordtune became the first product

08:14 – From B2C success to a pivot back into enterprise

09:43 – What AI21 learned from Wordtune for enterprise AI

11:15 – Defining “product algo fit”

12:27 – Training models before it was cool: Jurassic, Jamba, and beyond

13:38 – How to hire model-training engineers with no playbook

14:53 – Recruiting systems talent: what to look for

16:29 – How to orient your models around real enterprise needs

17:10 – Why Jamba was designed for long-context enterprise use cases

19:52 – What’s special about the Mamba + Transformer hybrid architecture

22:46 – Experimentation, ablations, and finding the right architecture

25:27 – Bringing Jamba to market: what enterprises actually care about

29:26 – The state of enterprise AI readiness in 2023 → 2025

31:41 – The biggest challenge: evaluation systems

32:10 – What most teams get wrong about evals

33:45 – Architecting reliable, non-deterministic systems

34:53 – What is Maestro and why build it now?

36:02 – Replacing “prompt and pray” with AI for AI systems

38:43 – Building interpretable and explicit agentic systems

41:09 – Balancing control and flexibility in orchestration

43:36 – What enterprise AI might actually look like in 5 years

47:03 – Why Israel is a global powerhouse for AI

49:44 – How Ori has evolved as a leader under extreme volatility

52:26 – Staying true to your mission through chaos

Subscribe to the Barrchives newsletter: https://www.barrchives.com/

Spotify: https://open.spotify.com/show/37O8Pb0LgqpqTXo2GZiPXf

Apple: https://podcasts.apple.com/us/podcast/barrchives/id1774292613

Twitter: https://x.com/barrnanas

LinkedIn: https://www.linkedin.com/in/barryaron/

What does it take to reimagine the browser—one of the most commoditized technologies in the world—for the enterprise?

In this episode, Ofer Ben Noon, founder of Talon and now part of Palo Alto Networks, shares the wild journey from exploring digital health to building the world’s first enterprise-grade secure browser.

We dig into:

• Why the browser became the new security perimeter

• How Talon raised a $26M seed and scaled fast

• What it takes to compile Chromium daily (and why it’s so hard)

• Why Precision AI is essential to secure AI usage in the enterprise

• And how generative AI, SaaS sprawl, and autonomous agents are reshaping enterprise risk in real time

If you care about AI x cybersecurity, endpoint security, or enterprise infrastructure—this is a deep, real, and tactical look behind the curtain.

This episode is broken down into the following chapters:

00:00 – Intro

01:05 – Why Ofer originally wanted to build in digital health

02:15 – The pandemic shift to SaaS, hybrid work, and browser-first

04:44 – Why Chromium was the perfect technical unlock

05:27 – The insane complexity of compiling Chromium

07:10 – What makes an enterprise browser different from a consumer browser

09:36 – Browser isolation, web security, and file security

10:50 – Why Talon needed a massive seed round from day one

11:53 – What an MVP looked like for Talon

14:08 – Early skepticism from CISOs and how Talon earned trust

16:50 – Discovering new enterprise use cases over time

17:11 – How AI and Precision AI power Talon’s security engine

19:21 – Why Ofer chose to sell to Palo Alto Networks

21:06 – Petabytes of data, 30B+ attacks blocked daily

23:44 – The risks of LLMs and generative AI in the browser

24:24 – What Talon sees when users interact with AI tools

25:05 – The #1 risk: privacy and user error

26:43 – Why AI use must be governed like any other SaaS

27:22 – How Talon built secure enterprise access to ChatGPT

28:05 – Mapping 1,000+ GenAI tools and classifying risk

29:43 – Real-time blocking, DLP, and prompt visibility

31:25 – Why user mistakes are accelerating in the age of agents

32:04 – How autonomous AI agents amplify risk across the enterprise

33:55 – The browser as the new control layer for users and AI

36:57 – What AI is unlocking in cybersecurity orgs

39:36 – Why data volume will determine which security companies win

40:28 – Ofer’s leadership philosophy and staying grounded post-acquisition

42:40 – Closing reflections

Subscribe to the Barrchives newsletter: https://www.barrchives.com/

Spotify: https://open.spotify.com/show/37O8Pb0LgqpqTXo2GZiPXf

Apple: https://podcasts.apple.com/us/podcast/barrchives/id1774292613

Twitter: https://x.com/barrnanas

LinkedIn: https://www.linkedin.com/in/barryaron/

Vanta helped create the automated security and compliance category—and now, they’re redefining it with AI.

In this episode, Christina Cacioppo (CEO & Co-Founder) and Iccha Sethi (VP of Engineering) join Barr Yaron to go deep on how AI is transforming the way Vanta builds products, evaluates models, and helps companies earn and demonstrate trust.

They cover:

- Why compliance is the perfect playground for AI

- How Vanta balances reliability, explainability, and scale

- What it takes to build golden datasets in high-stakes domains

- The real-world AI infrastructure behind Vanta AI

If you care about real AI product development—not just hype—this is a masterclass in doing it right.

🔔 Subscribe for more deep dives with leading AI builders and thinkers.

This episode is broken down into the following chapters:

00:00 – Intro

01:06 – Christina’s early entrepreneurial roots (Beanie Babies & all)

02:51 – From venture to founder: why Christina started Vanta

04:00 – What Vanta actually does

05:32 – Iccha on why she joined as VP of Engineering

07:09 – When Vanta started leaning into AI

08:33 – AI’s growing role in Vanta’s product roadmap

09:52 – How AI powers questionnaire automation

12:25 – Using LLMs to map policy docs to cloud configs

13:27 – Building trust: human-in-the-loop and explainability

16:03 – Vanta’s evaluation system for AI features

18:17 – How golden datasets are constructed (and maintained)

20:59 – Feedback loops: online eval from user behavior

22:43 – How model feedback informs product updates

23:38 – What Vanta wants from foundation models (but isn’t getting yet)

24:32 – Retrieval: how Vanta processes customer documents

27:13 – The hardest technical challenges in AI integration

29:41 – Internal adoption: how non-technical teams are using AI too

31:52 – Vanta’s centralized AI team & how other teams plug in

33:27 – Internal education: building AI intuition org-wide

34:31 – From prototype to production: experimentation culture

36:41 – Customer sentiment around AI in compliance workflows

38:22 – Enterprise buyers & the AI “kill switch”

39:06 – Personalized experiences as the future of trust

40:21 – How enterprises are approaching AI risk assessments

41:50 – What excites Iccha and Christina about the future of AI at Vanta

Subscribe to the Barrchives newsletter:

https://www.barrchives.com

Spotify:

https://open.spotify.com/show/37O8Pb0LgqpqTXo2GZiPXf

Apple:

https://podcasts.apple.com/us/podcast/barrchives/id1774292613

Twitter:

https://x.com/barrnanas

LinkedIn: https://www.linkedin.com/in/barryaron/

What if AI could talk back instantly—and naturally?In this episode, Karan Goel, Co-founder & CEO of Cartesia, joins Barr Yaron to unpack the future of voice AI, state space models (SSMs), and why audio is the next frontier in AI.

Karan shares the founding story behind Cartesia, explains how alternate architectures like Mamba enable ultra-efficient, low-latency inference, and walks through how his team is building the fastest text-to-speech model in the world—while obsessing over every millisecond.

Whether you’re into model architectures, AI infrastructure, or the future of voice interfaces, this episode delivers technical depth, startup lessons, and a roadmap for what’s coming next.This episode is broken down into the following chapters:

00:00 – Intro

01:06 – Karan’s journey from CMU PhD to startup founder

03:56 – Why Cartesia is built around state space models

06:49 – What makes SSMs different from transformers

09:14 – Why compression matters for long-running AI systems

11:13 – What data types SSMs are best (and worst) for

13:39 – Scaling SSMs: What’s possible and what’s missing

15:31 – Hardware, GPUs & why SSMs work well on existing infra

18:46 – Landing on audio: Cartesia’s first core modality

22:38 – Navigating the model vs. market debate in AI startups

26:36 – How Cartesia built Sonic, their ultra-low latency TTS model

28:17 – Why latency is the #1 challenge in voice AI

30:46 – Tricks vs. model-first thinking: Baking it into the model

34:01 – How Cartesia balances fast execution with deep research

36:26 – Building with part-time academic co-founders

38:13 – Yes, every employee gets a personal Yoshi

40:02 – Where voice AI is being adopted first (telephony + beyond)

42:24 – Multilingual modeling & the long tail of language

45:02 – Voice as a new computing interface

46:26 – Why voice notes are the future (and Barr’s hot take)

49:56 – How Cartesia evaluates its models

52:44 – How Karan has grown as a founder and leader

Subscribe to the Barrchives newsletter: https://www.barrchives.com/

Spotify: https://open.spotify.com/show/37O8Pb0LgqpqTXo2GZiPXf

Apple: https://podcasts.apple.com/us/podcast/barrchives/id1774292613

Twitter: https://x.com/barrnanas

LinkedIn: https://www.linkedin.com/in/barryaron/

AI is changing the game for marketing, but how do businesses truly integrate AI into their workflows?

In this episode, Kashish Gupta, Co-founder & Co-CEO of High Touch, joins Barr Yaron to break down the rise of Composable CDPs, the role of AI decisioning, and the future of automated marketing.

They explore:

- How High Touch pivoted into the Composable CDP space

- Why marketers struggle with data democratization and AI adoption

- How reinforcement learning (RL) agents are optimizing marketing campaigns

- What AI can (and can’t) replace in modern marketing teams

- The biggest technical challenges in building AI decisioning systemsIf you’re a marketer, data leader, or AI enthusiast, this episode is packed with insights into the future of AI-powered marketing.

This episode is broken down into the following chapters:

00:00 – Welcome & Introduction

01:09 – The High Touch founding story & early pivots

02:28 – Why Composable CDPs are the future of data-driven marketing

03:42 – Overcoming market education challenges

05:08 – The manual way of marketing vs. AI-powered decisioning

09:21 – How AI decisioning works inside High Touch

12:24 – Why reinforcement learning (RL) is the right AI model for marketing

13:45 – How High Touch sets up reward functions for AI agents

16:04 – How much data do you actually need for AI-driven marketing?

17:27 – The biggest misconceptions about AI and marketing

21:00 – Building an AI team at High Touch & hiring strategy

24:27 – The biggest technical challenges of AI decisioning

27:10 – How customers interact with AI-driven marketing models

29:40 – What’s next for AI decisioning & autonomous marketing agents?

34:26 – The future of AI & CDPs – where is the industry headed?

37:42 – What remains uniquely human in marketing?

41:23 – Advice for marketers considering AI adoption

44:16 – The future of AI agents working together

Subscribe to the Barrchives newsletter: https://www.barrchives.com/

Spotify: https://open.spotify.com/show/37O8Pb0LgqpqTXo2GZiPXf

Apple: https://podcasts.apple.com/us/podcast/barrchives/id1774292613

Twitter: https://x.com/barrnanas

LinkedIn: https://www.linkedin.com/in/barryaron/