David Singleton

The hottest role in AI startups right now isn’t Forward Deployed Engineers. It isn't GTM Engineers. It’s Deployment Strategists. Decagon calls it an “Agent Product Manager.” Harvey calls it a “Solutions Architect.” Palantir Technologies has had versions of this role for years. And the salaries are climbing fast: - Decagon: $200k–$285k - Palantir Technologies: $120k–$200k - Figma: $150k–$260k - Ramp: $100k–$180k - Harvey: $190k–$260k So who are these people? They are usually pseudo-technical -- CS or engineering majors, or folks with technical work experience. Many come from 2 years in consulting, IB, or PE, then jump into startups to get their hands dirty. They are young, hungry, polished, and comfortable being in front of customers. What do they actually do? They make sure enterprise AI deployments succeed. A $100k+ deal does not survive on a nice pitch or a self-serve onboarding flow. It survives if the customer sees value in the pilot. That means: - Embedding directly with the customer - Designing prompt logic for specific workflows - Working with engineering to align integrations and data flow - Helping exec teams define their AI roadmap - Running feedback loops into product and GTM Why does this role matter so much? Because enterprise AI is messy. Integrations, data transfer, and adoption make or break a deal. Most buyers are using AI for the first time, and each has unique workflows. Deployment Strategists bridge that gap. They own the outcome. They are accountable for making pilots successful, which often means millions in revenue down the line. At Conversion, Sam Bochner has been leading this work for us. We are now thinking about scaling it into a full team. Because a few successful pilots can fund an entire department, and the cost of failed deployments is too high to ignore. Is this just a rebrand of customer success? Not really. Success is about answering tickets and renewals. Deployment Strategy is about going deep with a few enterprise accounts, extracting maximum value, and ensuring the pilot closes into a multi-year contract. Call it Agent PM, Solutions Architect, or Deployment Strategist. Whatever the title, this is becoming one of the most important roles in AI SaaS.

654

91 Comments

GPUs have gotten faster, and models have gotten larger. But storage is still stuck in the past. Modern AI workloads need instant access to huge datasets. Instead, teams waste time on slow retrieval, idle GPUs, and costly pipelines. Archil is changing this. It delivers object-storage scalability at block-storage speeds. The platform allows for seamless, high-performance access to massive datasets. By eliminating cold starts and accelerating throughput 30X, Archil enables teams to fully utilize their infrastructure and move faster on training, analytics, and AI deployment. Archil’s founder Hunter Leath brings rare technical depth and operating experience from building Amazon EFS and optimizing Netflix’s cloud performance. This is a team building the foundational layer that AI workloads desperately need. I’m thrilled to lead Archil’s seed at Felicis with Nancy Wang and support their mission to reinvent cloud storage for the AI era. https://lnkd.in/gWueXhzz

65

2 Comments

There’s been a lot of discussions on the best way to build agents:  OpenAI's general take is that increased model capabilities simplify the SDK  Others (incl. Anthropic agent pattern guide + dexhorthy's 12 factor guide) generally outline a more constrained approach. We’ve been asked by a bunch of people on where we stand. High-level we agree with both: ✅ We agree with OpenAI that a lot of existing agent approaches are a bit cumbersome/overengineered to build agents. This is why we actually moved from a declarative graph-based approach in LlamaIndex (our old QueryPipelines) to workflows as an event-driven pattern. Also as models get better inevitably you will trust general agent reasoning more. Not everyone wants or needs to build agents from scratch. ✅ At the same time there’s a reason these agent design patterns exist. They are best practices for a good swath of use cases where you need to encode a specific business process. I personally really like Dex's 12 factor guide - it’s a collection of best practices that ensure you actually deeply understand the agent you’re building. As a framework, LlamaIndex workflows supports the entire spectrum, from constrained/from-scratch to general reasoning. The really nice thing about it is it doesn’t feel that different from just writing normal Python/Typescript, so you don’t have to learn a completely new design pattern, which makes it way more seamless to transition from regular programming. Huge shoutout to Laurie Voss for the blog Blog: https://lnkd.in/dVtv7Mgq For those getting started check out our workflows guide: https://lnkd.in/gRsUHuV8 12 factor guide coming in soon! 

254

14 Comments

SF AI-Native Operator Takeaway #2: In AI-native PLG, the hard part isn’t conversion... it’s discovery. Many AI-native teams are still talking about PLG using a classic SaaS mental model, but based on operator conversations in SF, that model is starting to break down in fairly obvious ways. The biggest bottleneck right now isn’t conversion. It’s discovery. In traditional PLG, users generally understood the category before they ever signed up. The problem was obvious, the product’s value was legible from the homepage, and the “aha” moment tended to show up quickly in first use. In that world, PLG meant optimizing onboarding, reducing friction, and improving free-to-paid conversion because user intent already existed. AI changes that assumption. In AI-native products, users are often curious but unclear. They don’t yet know what’s possible, value depends heavily on workflow, context, data, and role, and the product can feel abstract until it’s applied directly to their job. As a result, many users stall not because the product isn’t valuable, but because they haven’t discovered how it fits into their world and how they can't live without it. This is the real distinction people kept coming back to. PLG conversion answers, “Is this worth paying for?” PLG discovery answers, “What problem does this solve for me, right now?” What’s working best in practice is less about funnel polish and more about clarity up front: role- or workflow-specific entry points, guided examples instead of blank states, and opinionated first actions that show users a concrete outcome before asking them to explore. This also explains a broader pattern across AI-native companies. Forward-deployed teams and services-heavy delivery aren’t just implementation tools; they’re discovery mechanisms. They translate abstract AI capability into concrete workflow value, observe real use cases users wouldn’t self-discover, and feed those learnings back into what eventually becomes productized. PLG isn’t going away, but in AI-native companies it’s being redefined. Self-serve no longer means self-explanatory. Education becomes part of the product, and discovery has to come before optimization. The teams making progress aren’t obsessing over conversion rates yet. They’re focused on whether users see themselves in the product, how quickly they reach a meaningful outcome, and whether the product helps users get to a meaningful outcome for themselves quickly, without too much guesswork. Bottom line: in AI, PLG is less about removing conversion friction early and much more about creating understanding first. Once they understand, they may be hooked. Tomorrow is my last SF AI operator takeaway focusing on everyone's favorite topic du jour: 996 work schedules.

15

1 Comment

As I'm leaving AWS re:Invent in Vegas, I’m reminded of a counterintuitive truth: competing with the public clouds is often really good business VCs love to ask, “Why wouldn’t the incumbent do this?” But in markets this large (and evolving this quickly) that’s rarely the right question. The real questions are: Who can assemble the best team, get to market fastest and with the most focus? And where are there opportunities to partner with the biggest players rather than compete? We’ve seen this pattern repeat across the ecosystem: -People assumed Vercel and AWS were destined to compete. Today, they’re strategic partners with CEO Guillermo Rauch in the keynote -Snowflake vs. Amazon Redshift. Now, AWS is one of Snowflake’s strongest distribution channels -In Datadog’s early pitches, the concern was, “If it’s just monitoring AWS, why not use CloudWatch?” Fast forward to now, and Datadog is now $50B+ public company We’re going to see the same dynamics play out with OpenAI and Anthropic. These are becoming foundational platforms. But that doesn’t preclude incredible companies from being built on, next to, and even around them. Some will look competitive. Many will end up as partners. And the biggest outcomes often start in those gray zones

217

26 Comments

2 years ago today, A16Z announced they put $100M into Pinecone, as everyone and their mother tripped over each other to invest in vector databases. Given how little standalone vector DBs are valued now, the hype cycle is wild to look back on. Quick timeline below: Late 2022: • ChatGPT launches. • Early RAG architectures emerge. • First buzz around Pinecone, Weaviate, Milvus, Chroma. Q1 2023: • Massive demand surge for vector search. • Vector DBs seen as critical infra. • Heavy VC excitement (“new database category!”). Q2 2023: • Peak hype. • Pinecone raises $100M, Weaviate raises $50M. • Every AI infra diagram has a vector DB layer. Q3 2023: • Saturation hits. • Postgres, Redis, MongoDB add native vector search. • Clouds bundle it. Q4 2023: • Hype fades. • Vector search becomes a commodity. Early 2024: • Market tightens. • Attention shifts to agents, fine-tuning, orchestration.

763

117 Comments

Reading Metronome’s Monetization Operating Model, I kept coming back to one idea: pricing has become product. Software now delivers outcomes, not access. Yet most companies still charge as if they’re selling seats or licenses. That disconnect creates friction: for customers, unpredictability; for companies, stalled growth. The paper’s argument is simple but sharp—monetization isn’t a late-stage decision. It’s strategic infrastructure. Pricing needs the same ownership and iteration as any feature. Treat it like a surface that customers touch, not a spreadsheet buried in finance. If value is continuous and dynamic, pricing must be as well. That means product, GTM, finance, and engineering working from one system of truth. How many of us still treat pricing as an afterthought—when it should be a growth engine? https://lnkd.in/gnH7WzYf #Monetization #ProductStrategy #AI

8

The big labs are expansive in their product ambition, especially since foundation models have largely improved in lockstep - in order to compete with them you have to do things they won’t which are: - building a very rich software ecosystem around a primitive - orchestration across multiple models - going insanely deep on product and growth for a narrow vertical domain

58

3 Comments

fresh off the a16z press (link in comments)... We are at an interesting crossroads in what it takes to build Enterprise-focused software companies in the AI era. On one hand, companies are ramping faster - no longer is a 0 to 1M enough to raise your Series A a few years from launch. Companies are building with sub-5 member teams, and crossing previously unseen revenue benchmarks with extreme revenue efficiency and often profitability in a matter of months. On the other hand, fund sizes are becoming larger, Seed rounds are growing to 10s of millions and "early-stage" VCs (think non-Seed, multi-stage funds) are becoming conservative, wanting to see more traction before they fund. YC and other platforms are emerging as natural winners. How will the venture world look like 2 years from today? 🤔 (1) Will we see companies raising small rounds of a few million dollars, launch and then grow from their own profits, maybe raising very large rounds just to scale distribution, breach 100s of millions in revenue, and go for an IPO? (2) Or will we continue to see the typical Pre-Seed -> Seed -> A -> B -> C -> D -> E -> IPO journeys as the dust settles on AI prototyping and enterprises start measuring ROI on their AI spend? I'm all ears 👂

49

3 Comments

Meta’s Llama gets a home — and a strategy shift Meta just tapped Groq and Cerebras Systems (both private!) to host its open-source LLMs — but the real story might be Meta’s first real move into cloud AI. • The news: Meta partners with Groq + Cerebras to host Llama models, bypassing AWS, GCP, and Azure. • The strategy: Start with AI inference, add storage next? Meta might be quietly building an AWS competitor. • The play: Offer Llama hosting at-cost (or less) to win dev mindshare + undercut OpenAI. • Why it matters: Open-source + cheap inference = existential threat to OpenAI’s pricing model. • Trend watch: AI compute costs could race to zero — and Meta just joined the race.

31

1 Comment

I love the way Ozgun puts it: Ubicloud is an open-source alternative to AWS, at a fraction of the cost How do they dare take on these cloud giants? Özgun shares the backgrounds of the founding team (Daniel and Umur) and what it really takes to go head to head with big tech   More videos from e2vc Summit!

64

1 Comment

The real reckoning is the death knell of the 'non-venturing' VC. Most VC had turned into the 'first check' for ramen salary + first AWS check beyond the $10k credits. After that, opportunistic 'second check' on traction. If the gap between customer need and solution narrows significantly, thanks to GenAI, then that 'first check' isn't required. Entrepreneurs can keep tinkering, talking to customers, evolving that first solution (not MVP) and bank revenue once product/customer fit happens. If the 'first check' isn't required, the ability to write the 'second check' never appears. Thus, VCs end up having to take actual chances on _venture_ - i.e. funding crazy ideas that won't have revenue potential for years. If there is revenue potential, they won't be able to get in the startups early enough for it to matter. Banks will 'enter the chat' as they say in a big way since they are the best providers of capital when the revenue trajectory is clear and business de-risked. VCs that only invest in 'de-risked' startups weren't true VCs anyway. They will not be missed.

35

11 Comments

Just getting started with LLMs? Don’t let the jargon slow you down. This 5-part LLM Glossary is your shortcut to mastering the fundamentals👇 1️⃣ Model Types • Foundation Model: Pretrained on large-scale, diverse datasets to learn general capabilities. • Instruction-Tuned Model: Further trained to follow user instructions precisely. • Multi-modal Model: Handles multiple input/output formats like text, images, and audio. • Reasoning Model: Optimized for logic, problem-solving, and step-by-step thinking. • Small Language Model (SLM): Lightweight models for fast, efficient, on-device tasks. 2️⃣ Training & Fine-Tuning • Pretraining: Initial learning phase using massive, diverse data sources. • RLHF (Reinforcement Learning with Human Feedback): Aligns model behavior with human preferences. • DPO (Direct Preference Optimization): Uses ranked preferences instead of rewards for training. • Synthetic Data: AI-generated data used to supplement real datasets. • Fine-Tuning: Targeted retraining on specific domains or tasks. • LoRA / QLoRA: Techniques for efficient low-resource fine-tuning. • Guardrails: Rules or filters applied to enforce safety, ethics, and compliance. 3️⃣ Prompt Engineering • System/User Prompts: Define model behavior and user context in a session. • Chain-of-Thought (CoT): Prompts that guide the model to reason step-by-step. • Few-Shot / Zero-Shot Learning: Demonstrating tasks with few or no examples. • Prompt Tuning: Optimizing prompts via training for specific outcomes. • Context Window: The total number of tokens the model can "remember" per interaction. 4️⃣ Inference & Generation • Temperature: Adjusts output randomness; lower values are more deterministic. • Max Tokens: Limits the number of tokens in a generated response. • Seed: Controls reproducibility of outputs. • Latency: Time taken by the model to respond. • Hallucination: When the model generates factually incorrect but plausible-sounding information. 5️⃣ Retrieval-Augmented Generation (RAG) • Retrieval: Fetching relevant data from external sources before generation. • Semantic Search: Search based on meaning, not just keywords. • Embeddings: Vector representations of text used for similarity matching. • Chunks: Dividing documents into smaller segments for better retrieval. • Vector Databases (VectorDBs): Stores embeddings for fast and accurate search. • Reranking: Rescoring retrieved documents to prioritize relevance. • Indexing: Structuring data for efficient retrieval during generation. 🌟 Follow the AIKaDoctor (Free AI & Data Science Resources) channel on WhatsApp: https://lnkd.in/dCTCEKKc 📌Follow Habib Shaikh For more such content.

232

36 Comments

Greylock is leading the seed in SuperMe, the AI-native professional network. Most expertise is not yet legible to LLMs. Professionals demonstrate expertise in private contexts: product docs, meeting notes, and most often - in their own heads. SuperMe is building a way to extract and structure these insights into AI profiles, grounded in an expert's real words and personality. And it is network-first; the more conversations one's AI profile has, the more opportunities for experts to reinforce or create new insights -- expertise compounds. This is one of the strongest examples of founder-market fit we've seen at Greylock. Casey Winters is the most in-demand product & growth advisor that marketplace operators (self included) wish they could hire. And Ludo Antonov built & led engineering teams at some of the most iconic network businesses over the past decade: Whatnot, Lyft, and Pinterest (where he worked closely with Casey). The two started this company out of our offices, and have until now been quietly building a highly curated network with profiles like Elena, Lenny, Brian, etc. We see a unique opportunity to create the professional network graph in a world where likes & follower counts have lost meaning. More on our investment below:

286

13 Comments