Jeremiah Lowin

This section on how dbt Labs transitioned from a purely PLG motion to layering on enterprise is a fascinating one. Two instructive passages that I have bolded (below from First Round Review's path to PMF series featuring dbt Labs) TLDR: GTM comprises your ICP, channel, and message. When you transition from PLG / bottom up to Enterprise / top down motion, naturally your ICP and channel also changes, but your messaging / proposition needs to change too, e.g., the enterprise may be multiple personas and are buying assurance as much as solution. --- "Handy found product-market fit organically for dbt as an open-source tool mostly used by data practitioners and developers. But a few years into running a commercial business, he realized he had to build a growth curve all over again with C-suite data leaders. “Even though we had an unbelievable amount of market pull, as we initially commercialized, it wasn’t easy for us to transform this open-source command line tool into a product that enterprises would pay a million dollars for,” says Handy. “When you have enough product-market fit, sometimes it allows you to get away with not being super tight on product marketing or sales motions. So around 2022, we went from this gigantic acceleration curve and overnight we realized, we have to sit at the adults’ table and figure things out real fast,” says Handy. After the PLG flame started to fizzle, Handy turned his attention to layering on a sales-led motion for the cloud platform. “We had to focus our efforts on telling cohesive stories to senior data leaders. We had to have a very clear, explainable answer to the question, ‘Why should I use the commercial product and not the open-source product? And it had to be digestible by someone with a C in their title,” he says. Handy’s answer: dbt Cloud can handle complex data for companies of every size. “The longer people used dbt, the more complex their code became,” he says. “It was a problem for the most sophisticated dbt users, who were often at the largest companies. So there was a real opportunity for us to step in and solve that for them with dbt Cloud.” To tell that story to enterprise customers, Handy relied on data, naturally. “At a user conference we presented a chart that showed the number of dbt projects that had a certain number of models in them — over 100, over 1,000, et cetera,” he says. “We watched that number climb and we knew as users ourselves, ‘Oh my God, trying to work in a dbt project with 5,000 models in it is challenging.’ So we started with that quantitative data point and asked folks in our community about their experiences with these very large, complex dbt projects, and validated that this was a pain in the ass without a cloud platform.”

23

4 Comments

⚠️ Founders - consider the below reasons to not use Forward Deployed Engineers (FDEs) to acquire your *first* customers. This signals (1) lack of PMF; (2) miscalculation b/w development and sales timeframes; and (3) mischaracterization of Palantir Technologies timing when they deployed their FDE model. 💡 At the early-stage, it's crucial to demonstrate ability to repeatedly acquire customers with limited resources - capital, team, & time - that supports the narrative of 'more funding will amplify my traction.' Bespoke solutions for one company demonstrates zero repeatability. These custom solutions will not carry over to other customers within the same cohort. 💡 AI accelerates engineering but sales, system integration, and stakeholder buy-ins are perpetually slow. This imbalance results in concentrating all your engineering resources into one company, ignoring other opportunities, and still not going live on an accelerated timeline. 💡 Palantir didn't start using FDEs until year 3 when they got their first *gov't* customer, not their first customer. Gotham was already built and had strong PMF. Gotham allowed Palantir to take GTM risk by using FDEs b/c if it didn't work, Gotham was humming in the background. TLDR: At the early-stage, repeatable customer acquisition is the holy grail. Don't let AI deceive you that it can accelerate human behavior like it has with development cycles. It C.A.N.N.O.T.

33

1 Comment

In working with LLMs, a consistent source of performance gaps comes from limitations in memory and learning over time. In my own work, this gap shows up repeatedly: amazing performance in isolated tasks, followed by brittleness once interactions extend over time or require the system to build on prior experience. Memory and self-learning may not be the only way to frame the problem, but they are currently useful lenses for understanding where today’s LLM architectures break down. What makes this gap interesting is the contrast: these systems can be super-human in narrow capabilities, yet frustratingly wrong when even modest persistence or learning over time is expected. You don’t need to be a cognitive scientist to notice that the human brain is not just a very large context window. From a practical standpoint, it can be useful to think architecturally in terms of adding or improving memory and learning—not as the only way to frame the problem or its solution, but as architecturally useful working models. Seen this way, the issue is not access to more text via larger context windows, but the absence of persistent, updateable processes—mechanisms that govern what is retained, revised, and abstracted across interactions. I recently looked at Hindsight (link to repo in the comments), an open approach to agent memory that treats memory as more than a flat retrieval layer. Instead of storing everything in a single vector space, it introduces explicit memory categories that resemble familiar cognitive distinctions: 🔹World knowledge — relatively stable facts 🔹Experiences — what the agent itself has done and observed 🔹Opinions — beliefs that carry confidence or uncertainty 🔹Observations — higher-level models formed by reflecting across prior memory What stands out is not just the taxonomy, but the separation of memory operations: 🔹Retain information intentionally 🔹Recall relevant past context when needed 🔹Reflect to synthesize new insights from accumulated history This structure makes it easier to reason about how an agent’s knowledge evolves—particularly how beliefs change, how experience accumulates, and how abstractions form over longer horizons. Many existing RAG-style systems struggle here once interactions extend beyond short-term context. The approach has also been evaluated on LongMemEval, where it reports leading accuracy among open systems, with independent verification by external research groups. Whether or not this specific architecture becomes standard, it points toward a direction that feels necessary: systems that can accumulate experience, revise beliefs, and build internal models over time. That seems like a prerequisite for anything we would reasonably call self-learning. How are you thinking about memory and persistent learning in your own AI or LLM projects? And how much improvement do you think this could bring to LLM or AI systems’ performance? #AIAgents #LLMs #AgentMemory #MachineLearning #AIArchitecture

2

3 Comments

We've spent a lot of time thinking about where AI goes from here at White Star Capital. Current models are trained on text, but the world isn’t so binary. The hardest problems in robotics, industrial automation, and physical environments have nothing to do with predicting the next word… Which is why I'm excited to share that we're backing AMI - Advanced Machine Intelligence as part of its $1.03bn seed round. Led by Yann LeCun, Alex LeBrun, and with a team spanning Paris, Singapore, Montreal, and New York, AMI Labs is developing the next generation of AI systems beyond LLMs: World Models. World Models take a fundamentally different approach to LLMs. Rather than predicting the next token, they learn how systems evolve over time, how actions lead to outcomes, and how environments subsequently change. AMI Labs is building this as a general intelligence layer grounded in JEPA, the architecture Yann developed at Meta FAIR, trained on real-world multimodal data. The team speaks for itself. Chaired by Turing Award laureate Yann LeCun, led by CEO Alex LeBrun, with Saining Xie, Pascale Fung, Michael Rabbat, Laurent Solly, alongside a research team drawn from Meta FAIR, OpenAI and Google DeepMind. On a personal note, it’s an exciting moment to see a credible European alternative emerge, with strong links to Canada and Singapore, in a field that has been dominated by US Big Tech. For industries that require local data hosting, clear data ownership, and tighter control over where models are trained and deployed, AMI Labs shows great potential. White Star Capital | Matthieu Lattes | Bérénice Moustial | Victoria Pozzi Rocco Belforti More on the round here: https://lnkd.in/eiCKZKSQ

262

16 Comments

Why was the Fivetran-dbt merger all but inevitable? Fivetran & dbt Labs announced their merger yesterday. The all-stock deal combines two companies into an entity approaching $600 million in ARR. The beauty of the modern data stack was the explosion in choice. As the cloud exploded onto the scene, the legacy data warehouse was replaced by a collection of fast-moving platforms. In that era, specialization won. The pendulum is now swinging back towards consolidation. Why? The answer lies in compute economics & revenue scale asymmetry. The table below shows why. There are three different categories of software within this subset of the ecosystem: 1. Ingestion takes data from software & moves it into a cloud data warehouse. Snowflake acquired Datavolo, which commercializes the open source product Apache NiFi, calling it Openflow. Databricks acquired Arcion for ingestion through change data capture, calling it LakeFlow Connect. Fivetran focuses exclusively on this layer. 2. Transformation means reformatting the data within the cloud data warehouse. Snowflake launched native dbt Projects on Snowflake. Databricks offers Delta Live Tables, native SQL, & Python, plus supports hosted dbt through Databricks Workflows. dbt Core/Cloud is the leading independent transformation tool. 3. Compute revenue is generated when we ask questions of our data. Snowflake remains one of the leaders in structured data analysis with their cloud data warehouse. Databricks’ compute is their own as well. Here’s the asymmetry in one number. Compute represents 72% of the overall market ($7.6B of $10.6B). As a result of their massive operations, Snowflake & Databricks exert significant gravity within the ecosystem. They have expanded beyond the compute market to impose their presence & capture marginal revenue within customers, pressuring the competitive ecosystem. That’s not to say these components are independent. George Fraser analyzed Snowflake workloads in September 2024, finding transformation represents 40-45% of total Snowflake compute, which means even at smaller scales, startups can have significant impact on these behemoth businesses. The Fivetran-dbt merger is an inevitable evolution of a maturing market. Two unicorns must partner to compete against two decacorns. They solve two of the three customer problems. But not yet compute. One could surmise this consolidation signals the end of the modern data stack. I view it differently. The MDS has succeeded beyond our expectations. The stakes are higher now. Broad platforms, fast growth, & AI-native architectures define the next phase. Expect more consolidation. Category revenue estimates based on public disclosures & company filings. Ingestion: Informatica ($1.64B FY2024), Fivetran ($300M est.), Talend ($350M), others ($200M est.). Transformation: dbt Labs ($300M est.), others ($200M est.). Compute: Snowflake ($3.6B FY2025), Databricks ($4.0B ARR Aug 2025).

41

9 Comments

We’re excited to share that our fund has invested in Cerebras Systems. Artificial intelligence is entering a phase where infrastructure matters as much as algorithms. As AI adoption accelerates across industries, the demand for scalable compute to train and run increasingly complex models is growing exponentially. Cerebras is building technology designed specifically for this new era of AI. Their breakthrough approach centers on wafer-scale computing — a radically different architecture built to handle large-scale AI workloads with greater efficiency and speed. Instead of stitching together thousands of smaller processors, Cerebras developed a single, massive AI processor designed to simplify how large models are trained and deployed. What impressed us most about Cerebras is their bold rethinking of AI compute architecture and their focus on solving one of the most important challenges in the AI ecosystem: enabling organizations to run large AI models faster and at scale. As AI moves from experimentation to mission-critical enterprise deployment, the infrastructure powering these systems will become increasingly important. We believe Cerebras is well positioned to play a meaningful role in shaping that future. We’re excited to support the team as they continue to push the boundaries of AI infrastructure. #AI #ArtificialIntelligence #DeepTech #Infrastructure #Investing

26

3 Comments

Anthropic just shipped "auto mode" for Claude Code — a built-in classifier that blocks dangerous commands before they run. I built the same thing a few weeks ago because I got tired of Claude trying to shoot itself in the foot. Here's a sample from my logs of what a Haiku-powered permission hook has caught: - git reset --hard origin/main — "will discard all local commits without recovery" - git push origin --delete on every merged remote branch — all at once, no preview - A 40-line inline script calling delete_all on a critical table, mid-debugging session, no backup - Exposing an API token in plain text via curl (then logging it) - Running an arbitrary script from /tmp — "world-writable, origin unknown" The numbers after 9 days: → 2,106 commands checked → 217 blocked (~10%) → Total cost: $0.85 The AI wrote every single one of those blocked commands. I asked it to. It was trying to help. The problem isn't capability - it's that AI agents are enthusiastically literal. They do exactly what you asked, including the part where you didn't think through the consequences. Auto mode is the right instinct. "Check before you wreck" should be table stakes, not an afterthought. Welcome to the party, Anthropic. 🔒 https://lnkd.in/g26Khwp4

36

4 Comments