AWS Context [What It Changes]

TL;DR

• AWS Context is a preview service that maps a company's existing data into an organizational knowledge graph and gives agents identity-aware search over those relationships, as of the AWS Summit in New York on 17 June 2026.

• It is not a data warehouse replacement and not Amazon Quick's personal assistant graph; The Register reports Context as organizational rather than personal, with metadata published to Amazon S3 tables in Apache Iceberg format.

• No independent accuracy benchmark was published with the preview. Treat AWS's own description as the product claim, not a ranking.

• For a two-truck HVAC shop, a 10-person agency, or a solo clinic, the operational change is the same: agents stop guessing which spreadsheet is current, if and only if the shop already stores work in systems AWS Context can see.

Industry figures, not list prices.

Industry figures, not list prices.

Industry figures, not list prices.

According to AICPA, 62% of firms reported cloud-workflow adoption.

According to Journal of Accountancy, the mid-market close still runs 8-10 business days.

According to Thomson Reuters, tax-prep utilization hits 85-95% in March and April.

According to NFIB, 44% of small businesses cite time-management. According to SBA Office of Advocacy, 33M+ small businesses sit in the 2025 profile. According to Goldman Sachs, 62% of SMBs reported workflow-tool ROI inside 12 months.

Key Takeaways

• AWS Context is a knowledge layer for agents, not a new CRM and not a claim that warehouse tools are obsolete.

• The constraint that broke is agent context: models can plan, but they cannot safely traverse company relationships until those relationships are mapped and permissioned.

• Identity-aware queries inherit IAM and Lake Formation permissions; that is the governance story, not a third-party audit.

• Amazon Quick will use the same underlying graph technology; existing Quick users get the broader graph when Context is enabled.

• Preview status means there is no public accuracy score, no published SMB SKU, and no independent bake-off to cite.

What AWS Context is, in one sentence

AWS Context is a service that automatically maps relationships across a company's existing data into a knowledge graph and exposes agentic search so organizational agents can use governed relationships, business rules, and domain knowledge at runtime. That sentence is the entity. It is not a chatbot, not a replacement for Amazon S3, and not a claim that every small shop should migrate off QuickBooks.

A two-truck HVAC company already lives this failure. The dispatcher asks an agent which units are still under warranty; the answer comes from last year's export, not from the folder the office manager updated on Thursday. A 10-person marketing agency loses the same way when an agent drafts a client recap from the wrong drive. A solo-run clinic loses the same way when an agent cites an expired consent form. AWS Context is AWS's preview of a shared, governed map of those files, tables, and rules so the agent walks the real join path instead of a prompt guess.

That is why a shop that will never attend an AWS Summit still has a reason to read this page. The mechanism is organizational context with identity on every query. The product is a preview. The warehouse stays the warehouse.

Operators already routing intake into a CRM, the pattern in form-to-CRM automation, are looking at the same problem from the other end: capture is useless if the next agent cannot see which record is canonical. AWS Context sits after capture, not instead of it.

Why a small operator should care before the deep dive

According to the SBA Office of Advocacy, the United States contains 36.2 million U.S. small businesses. Almost none of those firms will staff a knowledge-graph team. They will still ask software to answer "which invoice is current" and "who approved that change."

According to the same SBA Office of Advocacy release, those firms account for almost 46 percent of private-sector employment and created a net 1.2 million jobs from March 2023 to March 2024. The Advocacy homepage restates the 36.2 million count and the 62.3 million small-business employees figure. That is the labor pool that cannot afford an agent that invents a join.

The NIST AI Risk Management Framework is the voluntary U.S. playbook for mapping, measuring, managing, and governing AI risk. AWS Context's identity-aware queries are one vendor's attempt to put access control on the query, which is the kind of control that framework asks operators to document. NIST does not certify AWS Context.

If the office already automates follow-up the way five ways to automate assistant tasks describes, Context is not a new receptionist. It is the map the receptionist-agent is allowed to read. Teams already routing invoices, work orders, and CRM notes through US Tech Automations workflows can treat AWS Context as a governed lookup step on that same path, not a rebuild of dispatch or billing.

A shop that wants the broader small-business picture, not the AWS SKU, can start from the state of small-business automation and come back here for the graph mechanics.

What AWS actually shipped on 17 June 2026

As of June 2026, AWS previewed AWS Context at the AWS Summit in New York. The summit roundup lists it under "Coming soon" and points to the context-intelligence post as the product write-up.

According to The Register, AWS previewed AWS Context on 17 June 2026 as a service for mapping company data into a knowledge graph for agentic search, organizational rather than personal, unlike Amazon Quick. The same article says Context publishes metadata into Amazon S3 tables in Apache Iceberg format and that queries are identity-aware.

The AWS product post matches that split. Mai-Lan Tomsen Bukovec's write-up says AWS Context extends the knowledge-graph technology that already powers Amazon Quick, where hundreds of thousands of users interact daily with a production graph that catalogs datasets, dashboards, and metadata and that already processes millions of requests per day. With Context, that personal graph becomes an organizational one. Glue Data Catalog, SageMaker Unified Studio, and Lake Formation integrate with the graph. Key elements publish to Amazon S3 in the Apache Iceberg format so customers can query metadata with Athena, Redshift, Spark, or any Iceberg-compatible engine.

That is the mechanism in plain language. The graph is a map of how tables, files, dashboards, and rules relate. Stewards review inferred relationships, promote them, and attach definitions. Agents query the map through agentic search APIs and MCP tools. Every query is supposed to inherit the caller's IAM and Lake Formation permissions so an agent cannot walk a path its user could not walk.

The same summit day shipped neighboring pieces that are not AWS Context but sit next to it: Amazon Bedrock AgentCore knowledge, web search, and paid-content paths; Amazon S3 annotations; Glue business context and semantic search in preview; Kiro for iOS in gated preview; Continuum security agents; DevOps Agent release management; and Transform continuous modernization. Context is the organizational graph. Those other launches are how agents get knowledge, how objects carry extra metadata, and how builders stay in the loop from a phone.

Sources: AWS Summit New York 2026 roundup; AWS Context post; The Register; S3 annotations; AgentCore; Kiro for iOS.

How the graph actually works

The failure mode AWS is naming is not "the model is too small." The AgentCore post says a capable model is only the starting point; agents fail when they cannot reach the SharePoint policy, the live web, or the paid feed, and when silent failures look like 99 percent dashboards hiding skipped approvals. Context is the organizational-knowledge half of that story. Managed Knowledge Base, Web Search, and WAF AI traffic monetization are the other layers on the same keynote day.

Stewards do not write SPARQL. They use a console to review inferred relationships, promote them to production, and attach business definitions and usage rules. As agents query the graph, AWS says it observes which sources produce correct results, which join paths get used, and which curated rules apply, then ranks sources by usage and shares what it learns across the organization. That is a vendor learning claim, not a published precision metric.

Portability is the Iceberg export. Apache Iceberg is an open table format for analytic datasets, with schema evolution, hidden partitioning, time travel, and compaction that Spark, Trino, Flink, and others can share. AWS Context publishes key metadata from structured and unstructured sources into Iceberg tables in Amazon S3 Tables. The shop that already queries S3 with Athena can audit the graph the same way it audits a fact table. The shop that does not run AWS still needs to know the format is not a proprietary lock on the metadata dump.

Governance is identity. Each call is designed to inherit the calling user's IAM and Lake Formation permissions. AWS Glue is the serverless catalog under many of those tables; Glue's Data Catalog and SageMaker Unified Studio are systems the Context post says integrate with the graph. Glue's own page states it connects to more than 100 data sources. That is Glue's connector count, not Context's coverage.

S3 annotations are the object-level cousin, generally available the same week. According to the S3 annotations announcement, each object can carry 1 GB of annotations per S3 object (up to 1,000 named fields, 1 MB each) in JSON, XML, YAML, or plain text. User-defined metadata remains 2 KB and immutable at upload; object tags remain 10 tags. Annotations are mutable, copy and replicate with the object, and flow into Iceberg tables when S3 Metadata is enabled, with refresh "within an hour" versus near-real-time journal tables. Query them with Amazon Athena or the S3 Tables MCP server.

Source: Amazon S3 annotations.

AgentCore's optimization loop — failure, intent, and trajectory insights, recommendations, and A/B tests — is the "did the agent actually do the thing" layer. Context does not replace that loop. An agent can query a perfect graph and still confirm an order it never placed. The Register quotes AWS on that class of silent failure.

Kiro for iOS is a native phone surface (chat, spec, autonomy; iOS 26+, gated preview). It does not host the Context graph. Agent Plugins 1.0.0 packages skills and MCP servers; AWS also lists the Agent Client Protocol — documented by JetBrains and Zed — as an editor-to-agent wire. Context is how an agent talks to company data, not how an IDE talks to an agent. The ACP GitHub repo states the stable protocol version is 1.

USTA analysis: annotation headroom versus header metadata

USTA analysis. Working only from figures already cited above: S3 user-defined metadata at 2 KB, annotations at up to 1 GB per object, and 1,000 named annotations of up to 1 MB each.

• 1 GB is 1,048,576 KB in binary kilobytes. Divided by the 2 KB user-defined metadata cap, that is 524,288× more capacity per object for annotations than for upload-time headers (1,048,576 ÷ 2).

• 1,000 × 1 MB is the advertised ceiling that produces the 1 GB total. That is a product limit, not a recommendation to store 1 GB of notes on every invoice PDF.

• This arithmetic says nothing about AWS Context graph quality. It only shows why AWS can tell agents to read object context without a sidecar database: the new slot is hundreds of thousands of times larger than the old header.

Sources for inputs: S3 annotations announcement. Results are USTA arithmetic, not an AWS benchmark.

Do not treat 524,288× as "the graph is that much smarter." It is storage headroom on the object. Context's graph still has no public precision, recall, or join-error rate.

What this does not establish

AWS Context does not establish that a two-person shop should rip out a file server. It does not establish a published price. The Register says Quick is subscription-based and DevOps Agent is per-second, with opaque task duration and extra charges for services the agent consumes. Context itself has no dollar figure in the sources fetched for this page. Check current pricing on the vendor site.

It does not establish independent accuracy. There is no third-party bake-off in The Register piece, the summit roundup, or the Context post. "Coming soon" is not general availability.

It does not establish that identity-aware queries satisfy a specific regulator. The NIST AI RMF is voluntary. According to NIST, AI RMF 1.0 was released on 26 January 2023, with a generative-AI profile (NIST-AI-600-1) on 26 July 2024 and a critical-infrastructure concept note on 7 April 2026. The PDF is NIST.AI.100-1; the AI Resource Center hosts the playbook. Those dates are NIST's. They are not an AWS Context certification.

It does not replace Apache Iceberg itself, AWS Glue, Lake Formation, or a warehouse. It maps relationships and publishes metadata. The data stays where it was.

A buyer's evaluation sequence

None of the sourced coverage says AWS Context files a warranty claim or books a truck. Use four checkpoints.

Silent-failure review still belongs in AgentCore insights and A/B tests, not in the graph console alone. If the HVAC shop's agent confirms a part order it never placed, that is the 99 percent dashboard problem AWS already named, not a missing node in the graph.

A startup that wants a generic agentic path rather than an AWS-only graph can look at agentic workflows and data-extraction agents as the parallel, then compare. US Tech Automations is the logging and approval layer if Context returns a join that still needs a person before anyone invoices the customer.

Pricing on this site is the plan surface for that logging layer, not an AWS quote.

Signal vs Speculation

Demonstrated signal: as of 17 June 2026, AWS previewed AWS Context at the New York Summit; The Register independently described it as an organizational knowledge graph for agentic search, distinct from personal Amazon Quick search, with Iceberg metadata in S3 tables and identity-aware queries; AWS's own post says the graph extends Quick's technology, integrates Glue / SageMaker Unified Studio / Lake Formation, and is queryable through agentic search APIs and MCP; S3 annotations GA the same week at 1,000 × 1 MB (1 GB) per object; SBA Advocacy counts 36.2 million U.S. small businesses; NIST's AI RMF remains the voluntary U.S. risk frame; no independent Context accuracy benchmark is in these sources.

Our read: over the next 12–36 months, small and mid-size operators will not "buy a knowledge graph." They will notice whether the agent that already drafts the recap, the dispatch note, or the invoice chase is reading the current folder or last year's export. If AWS Context stays a preview with no public accuracy number, the practical move for a 10-person shop is to tighten source-of-truth folders and identity on the tools it already runs, then treat Context as a model-and-map swap on an existing path rather than a platform migration. If identity-aware Iceberg metadata becomes a default on S3-heavy shops, the winners will be operators who already know which object is canonical — not operators who hope the graph invents it.

FAQ

What is AWS Context?

AWS Context is a preview AWS service that maps relationships across existing company data into an organizational knowledge graph and offers agentic search so agents can use governed relationships, business rules, and domain knowledge at runtime. It is not a personal Amazon Quick search index and not a warehouse replacement.

When was AWS Context announced?

AWS previewed it on 17 June 2026 at the AWS Summit in New York, as reported by The Register and listed as "Coming soon" in the summit roundup.

How is AWS Context different from Amazon Quick?

The Register and AWS both describe Quick's graph as personal and Context as organizational. When Context is enabled, Quick's agents can use the broader enterprise graph, including cross-system relationships and curated rules beyond one user's personal graph.

Does AWS Context replace a data warehouse?

No. It maps relationships and publishes key metadata to Iceberg tables in S3 so other engines can read the context. The warehouse, lake, and source systems remain the systems of record.

Is there a published accuracy score?

No independent accuracy benchmark is cited in the Register article, the summit roundup, or the AWS Context post. Do not treat the preview as a ranked retrieval product.

What should a small shop do this quarter?

Inventory which folders and tables an agent is allowed to see, put a person on any join that moves money, and keep capture-to-CRM paths intact. If the shop already runs AWS, watch the preview; if it does not, the lesson is still identity-aware context, not an AWS migration.

What to do next

AWS Context is a preview of a shared map for agents, with Iceberg metadata and identity on the query. The honest limit is the missing public accuracy number and the missing public price.

If the next step is to put that map behind an approval step instead of letting the agent invoice from a guessed join, start from agentic workflows for AWS Context-style lookups on US Tech Automations, then set the human hold on pricing.