# Sierra details Context Engine to turn customer history into an AI moat > Source: > Published: 2026-08-04 19:39:09+00:00 [Sierra](https://sierra.ai/about) on August 4th detailed Context Engine, a memory and decision layer designed to help its AI agents carry customer information across channels, identify behavioral signals and learn which actions produce business results. The release puts a product around Sierra's argument that enterprises can buy intelligence from changing model providers while retaining the customer context created through their own operations. [Bret Taylor (@btaylor)](https://x.com/btaylor), who co-founded Sierra with Clay Bavor, described that thesis as "rent the intelligence, own the relationship" [in a post on X](https://x.com/btaylor/status/2084716678202679596). Sierra's accompanying [product post](https://sierra.ai/blog/context-engine) says Context Engine collects information from customer interactions and systems holding purchase, billing, claims, appointment and usage records. It then connects those records to customers, products and transactions so an agent can decide what information matters at a particular moment. Taylor brings a direct view into both sides of that argument. He co-created Google Maps, served as Facebook's [chief technology officer](/article/rob-witoff-joins-coinbase-as-cto), founded Quip and later became Salesforce's co-CEO. He also serves on OpenAI's board. Bavor spent 18 years at Google, where his work included Google Workspace, Google Lens, Project Starline and Google Labs. Their bet with Sierra assigns commodity status to foundation-model access and puts durable enterprise value in the data and decisions surrounding each customer relationship. ### A memory layer above the models Context Engine is the data layer behind Sierra's [Horizon platform](https://sierra.ai/blog/horizon), which the San Francisco-based company introduced on July 16th. Horizon agents are designed to pursue goals over days, weeks or months, including originating loans, processing healthcare referrals, retaining subscribers and closing sales. That requires an agent to remember what happened between conversations and reassess its plan when a payment, appointment, product-usage event or customer response changes the situation. The Context Engine has three main functions. Its memory system ties interactions from different channels to the same customer and pulls information from systems of record. Its signal layer monitors events between conversations, such as a decline in product usage or an account change. Its personalization system selects a next action while operating within rules established by the customer, including limits intended to protect margins or customer-satisfaction scores. Sierra says the system also treats agent decisions as experiments. Most actions follow what has previously worked for customers in similar circumstances, while a smaller number test alternatives. An accepted retention offer, completed purchase or scheduled appointment then becomes evidence for future decisions. Over time, Sierra says Context Engine can use that outcome history to train models that predict churn, offer acceptance or conversion. That learning loop moves Sierra's agents beyond scripted support flows. A cancellation request, for example, can be evaluated against a subscriber's engagement, billing history, unused benefits and responses to earlier offers. The agent can then decide whether to offer a discount, recommend a different plan or avoid an incentive entirely. The design also raises the stakes for controls around automated experimentation. Sierra says customers define the outcomes and non-negotiable rules, while Context Engine chooses actions inside those boundaries. The product's value will depend on whether Sierra can show that those decisions produce measurable gains without allowing an optimization target, such as retention or sales, to override compliance requirements or customer interests. ### Sierra wants the context layer to compound Sierra's strategic claim is that foundation models deliver broadly available intelligence, while each enterprise's interaction and outcome history remains proprietary. A competitor can access the same model release. It cannot immediately reproduce years of evidence showing which offers, messages and sequences worked for a specific customer base. Sierra says customers retain control of the memory layer and can export it to a data warehouse or another external system through the [Context Engine product](https://sierra.ai/product/context-engine). That portability claim matters because the same accumulation that makes the data useful can make the platform holding it difficult to replace. Sierra is positioning itself as the system that organizes and acts on that context while arguing that the customer continues to own the underlying record. The economics follow the same strategy. Sierra says Horizon customers pay for completed business outcomes instead of model tokens. That structure transfers the immediate risk of inference costs to Sierra and gives Sierra an incentive to choose models, tools and workflows based on the cost of producing a result. It also places pressure on Sierra to define an outcome precisely, especially for processes such as retention, lending and healthcare that can stretch across many interactions. Sierra has ample capital to fund that work. On May 4th, Sierra [announced a $950 million round](https://sierra.ai/blog/better-customer-experiences-built-on-sierra) led by [Tiger Global](https://www.tigerglobal.com) and [GV](https://www.gv.com) at a valuation above $15 billion. [Axios reported](https://www.axios.com/pro/enterprise-software-deals/2026/05/04/sierra-ai-valuation-agentic-enterprise) a $15.8 billion post-money valuation. Sierra said in February that it had surpassed $150 million in annual recurring revenue, and in May claimed that its agents served more than 40% of the Fortune 50 and handled billions of customer interactions. Those operating figures remain company-reported. Context Engine shows where Taylor and Bavor intend to spend that credibility and capital. Sierra is moving from automating individual service conversations toward managing customer relationships across time. If the product works as described, the valuable asset is the growing record of decisions and outcomes surrounding each customer, with foundation models serving as replaceable components underneath it.