Sequoia Outlines Framework for Enterprises to Own Their AI Intelligence Stack

Sequoia Outlines Framework for Enterprises to Own Their AI Intelligence Stack Sequoia Capital details an emerging playbook for software companies to vertically integrate and own their domain-specific AI models, driven by advancing open weights and specialized post-training architectures. Industry leaders like Alex Karp and Satya Nadella have advocated for owning production means and reducing dependency on frontier labs. Open-weight models such as Kimi K3 and GLM 5.2 and independent tooling allow customized open models to rival frontier lab performance in specific domains. Decisions to move from closed APIs to custom models are driven by inference costs, execution speed, proprietary data protection, and product control. Technical execution requires a four-part pipeline: rigorous domain evals, custom harnesses, targeted post-training, and continuous online learning from production trajectories. While owning the stack raises engineering complexity and lowers the performance floor, it provides a higher performance ceiling than off-the-shelf frontier APIs. Strategic Drivers and Timing Advances in open-weight models like Kimi K3 and GLM 5.2 alongside tools from providers like Mercor and Fireworks enable companies to start near frontier baselines. Startups are shifting toward owning weights when rented intelligence creates bottlenecks around inference cost margins, latency demands in areas like coding autocomplete, data privacy, or the need to control product learning loops. Team Organization and Evaluation Benchmarks Transitioning to proprietary intelligence requires dedicated, small teams focused on building evaluations, tuning harnesses, and experimenting with models, rather than relying on general platform teams. Establishing repeatable evals is essential before model training; for instance, Harvey developed a Legal Agent Benchmark comprising over 1,200 tasks across 24 practice areas assessed by more than 75,000 rubric criteria. Harness Engineering and Post-Training Strategy The harness governs agent context, routing, tools, and traces, enabling task-specific routing and debugging. Post-training should adopt the lightest effective intervention according to Lin Qiao: context or RAG for factual gaps, supervised fine-tuning for formatting, preference tuning for product taste, reinforcement learning for specialized tasks, and distillation for latency or cost optimization. Online Learning and Tradeoffs Production experience is gathered through trajectories—tracking the context, tool calls, outputs, and user edits through platforms such as LangSmith to convert operational failures into new evals and harness fixes. Although owning the stack increases maintenance overhead and introduces a lower performance floor than turnkey closed models, it raises the product's ultimate ceiling.

红杉资本指南：企业如何掌控核心AI智能层 随着开源模型性能追赶前沿实验室以及后训练技术栈的成熟，红杉资本指出AI应用层的竞争正演变为对核心“智能层”的掌控之争，并为企业梳理了自建AI能力的战略考量与技术路径。 开源权重模型基线大幅提升且后训练工具生态成熟，使开源模型在特定垂直领域超越前沿闭源模型成为可能。 企业决定自建模型的核心驱动力包括降低推理成本、提升响应速度、保护专有数据壁垒以及掌控产品定义权。 技术实施路线图包含四个核心步骤：建立可量化评测体系、构建Agent线束与上下文工程、选择合适的后训练方法、建立在线学习与轨迹捕获闭环。 自建智能层虽增加了系统复杂度和下限风险，但也赋予了企业更高的能力上限，形成贴合具体业务的高效垂直模型。 自建智能层的成熟时机与动因 开源权重模型（如Kimi K3、GLM 5.2）的发展打破了过去微调即过时的死循环，加之独立后训练栈的完善，企业可直接获取接近前沿实验室的技术能力。企业在部分产品切片中自建AI的动力主要来自控制推理成本、在代码补全等场景追求极低延迟、确保专有数据不出内网，以及应对闭源实验室向产品层渗透所带来的竞争压力。 技术落地四大步骤 首先需建立评测体系（Evals），如Harvey针对法律场景构建包含1200个任务和7.5万条评分标准的基准；其次是搭建线束（Harness）负责路由、工具调用、记忆和上下文调度；第三是依据目标选用最轻量化的后训练手段（事实缺失用RAG、格式对齐用SFT、产品品味用偏好调优、专业任务用RL、降本提速用蒸馏）；最后通过捕捉任务轨迹（Trajectories）实现生产环境中的在线持续学习闭环。 自建路线的权衡与前景 直接调用前沿模型API虽然简单且底线高，但上限受限；自建开源模型、定制线束和私有评测体系虽然工作量庞大且面临更低的底线风险，却能大幅拉高能力天花板。未来的发展趋势是企业在前沿通用大模型之外，结合自身专有业务场景培养快速、专注且垂直的自有AI模型。

Published August 19, 2026 By , The race for the AI application layer is not only about UI, workflows, or GTM... it is a fight for the intelligence layer itself. The race for the AI application layer is not only about UI, workflows, or GTM... it is a fight for the intelligence layer itself. Published Aug 19, 2026 By By , The calls for owning your intelligence are escalating… Recently, Alex Karp encouraged enterprises to “own the means of production.” Not long after, Satya Nadella declared that buying intelligence from a frontier lab is like paying twice: once with money, and once with the proprietary knowledge you reveal to make that intelligence useful. The question everyone is asking: who should own the intelligence at the core of your business? To be clear, owning your intelligence is not a mandate nor a recommendation to get off the labs. For many workloads, frontier APIs and agents are still the right answer. But across our portfolio, we are seeing more companies building their own AI capabilities for parts of their products, vertically integrating toward owning and shaping their weights. A few weeks ago, Sequoia brought together a group of AI founders and builders for an event on owning your AI stack. We had Harvey give us the customer-back view. Then Mercor, Langchain, Trajectory, and Fireworks walked us through the technical building blocks in the post-training stack. Across the talks, a clear playbook emerged. Why now? First, the open-weight frontier has moved faster than anticipated. Kimi K3 and GLM 5.2 are extremely good! Training on open models used to be a treadmill… you might’ve spent months fine-tuning, and the next frontier release would erase your gains. Now, you can start from a baseline close to the frontier. Second, the independent post-training stack has matured. Thanks to companies like Mercor and Fireworks, every company can access the technology stack of a frontier research lab, from training to inference to human or synthetic data. With strong evals, harness engineering, post-training, and online learning, open models can now beat frontier models in specific domains. A year ago, you chose open weights when performance was something you could sacrifice. Now, it is becoming an existential and strategic question. The newest battleground is for the intelligence layer. Not your weights, not your product. When should you make the move? There’s no one-size-fits-all approach. In addition to performance, there may be slices of your product where renting intelligence becomes a constraint because of… 1. Cost. The more successful your AI product, the higher your AI COGS. If inference costs scale directly with usage, owning the model is the best way to protect your margins. 2. Speed. If you’re operating in a domain like coding (tab autocomplete) or cybersecurity, a small distilled custom model might beat a large general one, because speed is so important. 3. Proprietary data. If your feedback, evals, customer interactions, or domain data are what make the system better, you might prefer that data stays within your walls. 4. Controlling your destiny. The app layer and the intelligence layer are starting to merge. The labs are moving up into product, and app companies are moving down into the training loops that define how the product thinks. Owning the product increasingly might mean controlling more of the learning loop and the intelligence itself (h/t Harvey Research, RampLabs, Glean, Factory, and more). What’s the roadmap? Assuming you've decided on your strategy for what to own vs. rent, the question becomes: how do you go from zero to one? Build your team. Don't shoehorn this into a platform team. Owning your intelligence needs people playing offense: building evals, shaping data, experimenting with open models, tuning harnesses, and pushing performance in a domain. Small, de novo teams go very far here, partnering with half a dozen ecosystem partners. Harvey, for example, has done a surprising amount of research with a team of just seven people! Make the work legible. Every buyer right now is choosing their AI champion. Increasingly, the tiebreaker is published research, benchmarks, or technical posts. If you choose to do wonderful research in-house, share it with the ecosystem. Execute the technical steps. We’ve laid out a high level framework below. 1/ Evals Harvey’s Gabe Pereyra said it well: "If you don't have a good benchmark, you can't train models.” An eval is a set of tasks that measures whether your system can do the work well. Every task contains a prompt, context the model can use, and a grader. Most evals start as a founder squinting at outputs and vibe-checking whether they feel right. The goal is to turn that judgement into something repeatable. Harvey built its Legal Agent Benchmark by turning real legal work into discrete tasks a model could be tested on. The first iteration includes more than 1,200 agent tasks across 24 legal practice areas, and is evaluated by over 75,000 expert-written rubric criteria. It’s important to have an eval before making the decision to own your intelligence. Once every forward pass has an eval, deciding which model to use can be a measured decision instead of a guess. 2/ Harness and context engineering An agent has three parts: the model, the context, and the harness. The harness governs the product logic around the model: routing, retrieval, tools, memory, fallbacks, and traces. Harrison Chase’s simple framing was that "the main job of the harness is to bring context to the model at the right point in time." The more out of distribution the task is for the model, the worse the off-the-shelf harness will be. A good harness lets you route tasks to the best model for your specific job, reuse the same evals across models, and decide what context and tools an agent gets. It also makes the agent inspectable. You can trace what context went in, which tools were called, and where it got stuck. 3/ Post-training Post-training covers several techniques. The right one depends on what you’re trying to improve. Lin Qiao laid it out: if the model is missing facts, you don't need to post-train - just use context or RAG. If the output format or behavior is wrong, use supervised fine-tuning. If the issue is product taste, use preference tuning. If the model needs to improve at a specialized task, use RL. If the model is too slow or expensive, distill it. The goal is to pick the lightest method that moves your eval. Then serve the model through the same harness and measure quality, latency, and cost as it runs. 4/ Online learning Once the system has its pieces – evals, harness, and model – the final step is improving that system in production. Arjun Karanam made a good point: models keep getting smarter, but every session feels like it's their first day on the job. You can put Terence Tao in an accounting firm, but chances are (at least on day 1) he won’t be the best accountant there. What's missing is experience, not intelligence. Agents create that experience as they run. A trajectory is the path through the task: the context the model saw, the tools and sub-agents it called, the answer it produced, and what the user edited, undid, or retried. Capturing those trajectories in a tool like LangChain's LangSmith is necessary to build a continuous production loop. A failed task becomes an eval. Missing information goes into context or memory. A bad tool response becomes a harness fix. Final thoughts Buyer beware: owning your intelligence opens Pandora’s box. The closed-model stack is simple. You call a frontier model, use an out-of-the-box harness (e.g., Claude Code or Codex), add prompts and context, and ship. That gives you a high floor, but also a lower ceiling. Owning your stack means taking on more of the system yourself. The production stack becomes your open-source model, custom harness, tools, and context. The development stack becomes your proprietary evals, domain data, and online learning loop. It is certainly more work! It potentially gives you a lower floor – but it also raises the ceiling. The labs will keep building the giant brains. We should all use them. But, in parallel, the best product companies are growing their own little geniuses: fast, opinionated, domain-obsessed, and tuned to the work they see. In a world where more companies take ownership of their own intelligence, the ecosystem flourishes and individuality triumphs. That’s the promise of a world where companies own their intelligence. And we’re delighted to see the progress from the little guys, the startups that want to bring that world into existence. share