Life update: I've joined the ranks of the Substackers
I love creative humor writing, and until someone pays me to do it, I'll be publishing my work via this weekly newsletter.
I'd be quite flattered if you subscribed.
https://t.co/fXo0DjW3Rq
New on the blog today: What the Data Says About AI Pricing ๐
@Spyzguyz dug into our recent AI Pricing Report and shared some of his favorite takeaways.
๐ Read it here: https://t.co/ps13hjHaf8
AI is the most transformative technology of our lifetime.
But its scale is defined by electrons.
We just published our energy investment thesis โ 5โฃ technologies we believe will power the AI era.
A thread ๐งต
https://t.co/Q4XEVfNYxI
Announcing the Salesforce Ventures AI Pricing Report ๐ค
We surveyed 300+ startup and enterprise executives on AI pricing.
87% of sellers plan to change their pricing in the next 12-18 months.
Pricing is a massive, ongoing experiment. Here's what the data shows ๐งต
Our 2025 Year in Review is live ๐
โ 44 new investments
โ 30 follow-on investments
โ 20+ M&A exits
โ $1B AI Fund fully deployed
โ New focus areas in robotics, defense, & energy
Explore our full Year in Review: https://t.co/xHtEk7NGqX
X turned into a blogging platform and I caved.
In my defense, @AnthropicAI 's Series G and crossing $14B in run-rate revenue felt worth more than 280 characters, so it's for a good cause ๐
A lot of people donโt know this but @SalesforceVC has been an investor in Anthropic since the Series C back in early 2023 and weโve doubled down every single round including this most recent one. Many may think investing at the Series C was a no-brainer, but this was before neolabs raising at multi-billion valuations was a common occurrence. Anthropic is THE canonical success story that has paved the way for this archetype of startup journey, going from research to successful commercial products that can capture billions of dollars of value.
In honor of their fundraise, today we published a deep dive on our journey with Anthropic. My article with @pdrews unpacks the behind-the-scenes around how we made the decision to double down every single round.
In NYC: Dobbins Street Vintage, Dream Fishing Tackle, Lichen, tihngs, Humble House, Shop 86, Sterling Place
In HV: Newburgh Vintage Emporium (2 locations), The Antique Warehouse, Magic Hill Mercantile, Hyde Park Antiques Center + lots of small shops in Hudson, Kingston, Saugerties, etc.
Our final Salesforce Ventures portfolio session at #DF25 this year featured @Vercel CEO @rauchg and COO @jdewitt29 discussing their company's vision for the agentic web and software's next frontier.
Some highlights from the talk:
๐ ๐๐ก๐ ๐ฐ๐๐ ๐ข๐ฌ ๐๐ฎ๐ง๐๐๐ฆ๐๐ง๐ญ๐๐ฅ๐ฅ๐ฒ ๐ฌ๐ก๐ข๐๐ญ๐ข๐ง๐ ย โ Just as cloud re-made the web once, agents are poised to transform the web again with agents that can act on the users behalf. "You don't just want AI that responds, you want it to go work on a problem for three hours and then come back with an exact insight."
๐จโ๐ญ ๐๐๐ฏ๐๐ฅ๐จ๐ฉ๐๐ซ๐ฌ ๐๐ซ๐ ๐๐๐๐จ๐ฆ๐ข๐ง๐ ๐๐ ๐๐ง๐ญ ๐จ๐ซ๐๐ก๐๐ฌ๐ญ๐ซ๐๐ญ๐จ๐ซ๐ฌย โ "Our very competent devs are now managing 5-6 agents," Guillermo noted. "You're becoming an engineering manager even as an IC." Software engineering is evolving, and with AI democratizing development, Guillermo believes the number of people able to do engineering work will explode.
๐ค ๐๐ ๐๐ง๐ญ๐ฌ ๐๐ซ๐ ๐๐ฅ๐ซ๐๐๐๐ฒ ๐ฌ๐๐ฅ๐-๐ก๐๐๐ฅ๐ข๐ง๐ ๐ฌ๐จ๐๐ญ๐ฐ๐๐ซ๐ ๐๐ญ ๐๐๐ซ๐๐๐ฅย โ Guillermo said his company's agents now handle 75% of customer service tickets and are capable of responding to highly complex technical questions. When threats emerge, agents look into them first.
๐ก๐๐ฏ๐๐ซ๐ฒ๐จ๐ง๐ ๐ฌ๐ก๐จ๐ฎ๐ฅ๐ ๐๐ฎ๐ข๐ฅ๐ ๐ญ๐ก๐๐ข๐ซ ๐จ๐ฐ๐ง ๐๐ ๐๐ง๐ญ๐ฌย โ While agents-as-a-service will be powerful, Guillermo's advice is clear: "You're better off in the long run building your own, because you know your business better than anyone else."
A compelling vision for where software is headed. Thanks Guillermo and Jeanne for the insightsย and ending #DF25 on a strong note ๐ช
At #DF25's "Openness Fuels AI Democratization" panel, @percyliang, co-founder of @SalesforceVC portfolio company @togethercompute, shared his thoughts on why open source matters for the future of AI.
One of my favorite takeaways: Today's open-source AI feels like a similar trajectory as software in the 1990sย ๐
Back then,ย Microsoftย dominated and Linux was just a scrappy grassroots project. Now, Linux powers our entire internet ecosystem. Percy posed the question: "How can we build the Linux of foundation models?" He said that "there's a better way to build AI where everyone isn't a downstream consumer of AI. We can all shape AI for years to come."
Another sharp #DF25 conversation with a ton of good teachings. Thanks Percy for sharing ๐
.@SalesforceVC portfolio CEO Evan Smith (@Altana_AI) sat down this afternoon with Salesforce board member Amy Chang at #D25 for an engaging panel on how enterprises are navigating supply chain volatility ๐ข
Some conversation highlights:
๏ฟฝ๏ฟฝ ๐๐ก๐ ๐ฌ๐ฎ๐ฉ๐ฉ๐ฅ๐ฒ ๐๐ก๐๐ข๐ง ๐ข๐ฌ ๐ง๐จ๐ฐ ๐ ๐ง๐๐ญ๐ฐ๐จ๐ซ๐คย โ Evan likened Altana to LinkedIn'sย business relationship graph, but for the physical economy: able to track who owns what, who ships where, who builds what. In the modern world, you can't just manage buyer-supplier relationships anymore โ you need visibility into your suppliers' suppliers. Altana is helping enterprises gain this critical oversight.
๐ง ๐๐ฎ๐ฉ๐ฉ๐ฅ๐ฒ ๐๐ก๐๐ข๐ง๐ฌ ๐๐ซ๐ ๐๐๐ข๐ง๐ ๐ซ๐-๐ฆ๐๐๐ ๐ข๐ง ๐ซ๐๐๐ฅ ๐ญ๐ข๐ฆ๐ย โ "Supply chain network competition is the new normal, and it's unclear how it's going to play out," Evan said. "A resilient enterprise is one that is managing their business as an extended network and not just a series of buyer and supplier relationships." Enterprises are realizing the supply chains they've relied on aren't necessarily the ones they can rely on going forward โ and are now planning multiple contingencies in the face of increasing volatility.
๐ ๐๐ฎ๐ญ๐จ๐ฆ๐๐ญ๐ข๐จ๐ง ๐ข๐ฌ ๐ ๐ญ๐จ๐ฉ๐ข๐ ๐๐ญ ๐๐ฏ๐๐ซ๐ฒ ๐๐จ๐๐ซ๐ ๐ฆ๐๐๐ญ๐ข๐ง๏ฟฝ๏ฟฝ๏ฟฝ๏ฟฝ ๐ง๐จ๐ฐย โ But most enterprises are still trying to figure out where to start. Amy's advice: "We need to get into a spirit of experimentation. The more you can get your experimentation to tie to a bottom-line impact, the easier you'll find the change management required to make it happen."
A great conversation on a nuanced and important topic. Thanks Amy and Evan for your insights ๐
๐ At #DF25's "Rethinking Cloud Infrastructure for Engineers" panel,ย Salesforce Ventures portfolio founders Russ d'Sa (@livekit) and Alon Arvatz (@PointFive_Inc) chatted about what matters when it comes to AI infrastructure.
Some of my takeaways:
๐ง ๐๐ง๏ฟฝ๏ฟฝ๏ฟฝ๐๐ฅ๐ฅ๐ข๐ ๐๐ง๐๐ ๐ง๐๐๐๐ฌ ๐ข๐ง๐๐ซ๐๐ฌ๐ญ๐ซ๐ฎ๐๐ญ๐ฎ๐ซ๐, ๐ง๐จ๐ญ ๐ฃ๐ฎ๐ฌ๐ญ ๐ฆ๐จ๐๐๐ฅ๐ฌย โ AI companies talk about intelligence like it's a utility that you can just plug into, but even electricity needs power plants and charging stations to actually work. AI is the same way โ it needs a vast layer of infrastructure to be usable. And because the current internet wasn't designed for AI applications, to get voice and vision tools rolling, we need to rebuild the infrastructure from scratch. As Russ puts it: "Intelligence is also a raw material that requires a harness."
๐ค ๐๐ซ๐ฎ๐ฌ๐ญ ๐ข๐ฌ ๐ญ๐ก๐ ๐ซ๐๐๐ฅ ๐๐ ๐๐จ๐ญ๐ญ๐ฅ๐๐ง๐๐๐คย โ LLMs are black boxes: Users don't know what's happening behind the scenes. That blocks adoption. The fix comes from better observability and evaluation. Companies need to evaluate AI like they evaluate humans: test, verify, observe, repeat. Without that, enterprises won't roll it out broadly, no matter how good the technology gets.
๐ฐ ๐๐ ๐ฒ๐จ๐ฎ'๐ซ๐ ๐ง๐จ๐ญ ๐๐๐ซ๐๐๐ฎ๐ฅ, ๐๐ ๐๐๐ง ๐๐๐ญ๐ฎ๐๐ฅ๐ฅ๐ฒ ๐๐ ๐ฆ๐จ๐ซ๐ ๐๐ฑ๐ฉ๐๐ง๐ฌ๐ข๐ฏ๐ ๐ญ๐ก๐๐ง ๐๐ง ๐๐ฆ๐ฉ๐ฅ๐จ๐ฒ๐๐ โ AI agents can blow up costs fast if you're not running efficiently. That's why observability matters โ you need to know which model provider works best for your use case, and keep costs under control from Day 1.
๐โโ๏ธ ๐๐ง๐ญ๐๐ซ๐ฉ๐ซ๐ข๐ฌ๐๐ฌ ๐๐ซ๐ ๐ฆ๐จ๐ฏ๐ข๐ง๐ ๐๐๐ฌ๐ญ๐๐ซ ๐ญ๐ก๐๐ง ๐ฒ๐จ๐ฎ ๐ญ๐ก๐ข๐ง๐คย โ Fortune 500s are afraid of AI disruption, because they know that what they've built over decades could be replaced in just years with AI. That's pushing them to prioritize innovation instead of avoiding risks. They're getting more comfortable shipping AI-native products and competing directly with startups so they don't fall behind.
๐ง ๐๐ฏ๐๐ซ๐ฒ๐จ๐ง๐ ๐ข๐ฌ ๐๐๐๐จ๐ฆ๐ข๐ง๐ ๐ ๐๐ฎ๐ข๐ฅ๐๐๐ซ ๐ฐ๐ข๐ญ๐ก ๐๐ย โ"My marketing guy can build browser extensions using AI, and he's never written a line of code. He just vibe-coded it. Everyone in the org is a builder now." In startups where 60-70% of the staff are engineers, the ability for everyone to build is a welcome development. For enterprises, it's tough to govern but hard to avoid: It's the future of how work gets done.
Another sharp #DF25 conversation with a ton of good teachings. Thanks Alon and Russ for sharing!
๐ Salesforce Ventures portfolio founder @RichardSocher (@youdotcom) sat down with @OpenAI 'sย Head of Platform @oliviergodement and Salesforce's Adam Evans at #DF25 to discuss how AI is transforming enterprise work โ and what it takes to move from experimentation to implementation
Some of my key takeaways:
๐งฌ ๐๐ง๐ญ๐๐ซ๐ฉ๐ซ๐ข๐ฌ๐๐ฌ ๐ฐ๐๐ง๐ญ ๐ฏ๐๐ซ๐ญ๐ข๐๐๐ฅ ๐๐ย โ Fortune 500 companies don't want generic LLMs and experimentation. They want accuracy and ease of adoption. That means industry-specific AI that can easily train on an organization's data and solve for specific use cases that require unique context.
๐ ๐๐ก๐ "๐ฆ๐๐ง๐๐ ๐๐ซ ๐ฆ๐ข๐ง๐๐ฌ๐๐ญ" ๐๐๐ง ๐๐ ๐ ๐๐จ๐ฆ๐ฉ๐๐ญ๐ข๐ญ๐ข๐ฏ๐ ๐๐๐ฏ๐๐ง๐ญ๐๐ ๐ ๐ข๐ง ๐ญ๐ก๐ ๐๐ ๐๐ ๐ย โ Individual contributors need to learn delegation: giving precise context, verifying outputs, and building reliability with AI agents over time. It's part of fundamentally rethinking how knowledge workers operate. Being able to delegate and trust your AI agents will unlock new leverage for workers.
๐ป ๐๐ ๐๐จ๐๐ข๐ง๐ ๐๐ซ๐จ๐ค๐ ๐ญ๐ก๐ซ๐จ๐ฎ๐ ๐ก ๐ข๐ง 2024. ๐๐ฉ๐๐ซ๐๐ญ๐ข๐จ๐ง๐ฌ ๐ญ๐๐๐ฆ๐ฌ ๐๐ซ๐ ๐ง๐๐ฑ๐ญย โ Olivier said his engineering teams would "riot" if he took away their AI tools, because what used to take 6 months to build can now take 6 weeks. In 2025, we're hitting a similar inflection point with operations roles โ marketers and salespeople are using AI to build impactful tools because they understand their own workflows better than anyone else could.
๐ ๐๐๐ญ๐ ๐๐จ๐ฅ๐ฅ๐๐๐ญ๐ข๐จ๐ง ๐ฉ๐ซ๐๐๐ข๐๐ญ๐ฌ ๐๐ ๐๐ข๐ฌ๐ซ๐ฎ๐ฉ๐ญ๐ข๐จ๐งย โ Want to know how much AI will transform your industry? Look at how much data your industry collects. Digitized jobs with clear inputs and outputs are the lowest hanging fruit in the AI revolution.
Another banger hashtag#DF25 session ๐ฅ TY Richard, Olivier, and Adam for your insights.
The final day of #DF25 promises to be our busiest yet ๐โโ๏ธโโก๏ธ
You can find our team all day at the Agentic Theater in Moscone West for the following portfolio company sessions:
๐ค "How Generative AI is Transforming Work" with @RichardSocher of https://t.co/lnVnZGj7R7 alongside Salesforce's Adam Evans and @oliviergodement of @OpenAI.
โ "Rethinking Cloud Infrastructure for Engineers" with Alon Arvatz of @PointFive_Inc in conversation with @TIME's @byayeshajaved.
โป๏ธ "The Resilient Enterprise for a New Era of Trade" with Evan Smith of @Altana_AI in conversation with Amy Chang of Procter & Gamble.
โ "Openness Fuels AI Democratization" with @percyliang of @togethercompute and Salesforce's @JotyShafiq.
๐ "The Agentic Web Powered by Vercel" with Guillermo Rauch and Jeanne DeWitt Grosser of @vercel.
As usual, @Spyzguyz will be sharing live SFV updates from the conference, and download Salesforce Ventures' Guide to Dreamforce for curated content recommendations and advice for navigating #DF25: https://t.co/PQThzdd3lu
Let's close the week out strong ๐ช