AgentCore, One Year On
A year ago, integrating AWS Bedrock meant wiring InvokeModel calls into a Lambda, bolting on a vector store, and duct-taping a prompt chain together. Agent orchestration was something you wrote yourself. Session state was your problem. Tool invocation was your problem. Deployment was your problem.
That was the whole gap AgentCore was built to close — and with the April 2026 update, AWS has closed it aggressively. The biggest AgentCore release to date ships a managed harness that runs your agent the moment you define it, a CLI that makes agent deployment look like any other piece of cloud infrastructure, and a set of primitives (persistent filesystems, OS-level browser control, policy evaluations) that turn a 2025 'hello world' agent into something you could actually put in front of a customer.
The rest of this piece walks through what changed, why it matters, and what it means if you're a marketer looking at this stack and wondering whether you're supposed to operate it or ignore it. Spoiler: you're supposed to ignore it, and we'll explain why.
The Managed Agent Harness: First Working Agent in Minutes
The headline feature is the Managed Agent Harness (currently in preview). You define three things — a model, a system prompt, and a set of tools — and AgentCore runs the agent immediately. No orchestration code. No scaffolding. No bespoke FastAPI wrapper with a retry loop you copied off GitHub.
More importantly, every session gets its own microVM with a live filesystem and shell access. That's a genuinely new primitive. The agent can download files, run code, inspect its own outputs, and carry state across turns without you provisioning a single container. When paired with the new managed session storage (also in preview), the agent's filesystem persists — meaning an agent can suspend mid-task and resume exactly where it left off, files intact.
For context: building this yourself in 2025 required Firecracker, a homegrown session store, a custom checkpointing layer and a small team of platform engineers. AWS has now collapsed it into a managed service with no additional charge for the harness.
AgentCore CLI and Infrastructure-as-Code
The second big shift is governance. The new AgentCore CLI lets you deploy agents the way you deploy any other AWS resource — declaratively, version-controlled, reviewed in a pull request. AWS CDK support is live today; Terraform support is coming soon.
This matters more than it looks. The reason most enterprises refused to put agents in production through 2025 wasn't model quality — it was the absence of audit trails, change management and rollback. A prompt is a production artefact. You can't leave it in a Google Doc. The CLI treats agents as infrastructure, which means security, finance and compliance finally have something to point their existing tooling at.
Skills for Claude Code, Codex, Cursor and Kiro Power
AgentCore Skills is the feature that's going to move fastest culturally. Skills are reusable capability bundles — auth, retrieval, data access, tool chains — that coding assistants can pick up on demand. They're available in Kiro Power today, with Claude Code, Codex and Cursor support coming next week, per AWS's announcement.
The implication: developers working in any of the major AI-native IDEs will soon be able to call AgentCore Skills as first-class primitives. Your internal 'look up a customer in Salesforce' capability becomes a Skill. Your 'post to the campaign calendar' capability becomes a Skill. They ship once, they work everywhere, and they run inside AWS's policy perimeter.
Browser OS Actions and Filesystem Persistence
The AgentCore Browser got an upgrade that pushes it past headless Chromium territory: OS-level interaction capabilities. The browser agent can now act on the operating system itself — open applications, manipulate windows, trigger system dialogs. Combined with filesystem persistence, this is the first credible AWS answer to the 'agent that uses a computer like a person does' pattern that Anthropic's computer use and OpenAI's Operator kicked off.
AWS also added AG-UI protocol support (landed in March 2026), giving developers a universal interface for agent UIs — and quality evaluations and policy controls for what the announcement blog frames as 'trusted AI agents.' Translation: you can now evaluate agent behaviour against rubrics and gate deployments on policy checks, which is the last thing enterprise legal teams were waiting for.
Availability: Oregon, N. Virginia, Frankfurt and Sydney. No additional charge for the harness, CLI, or Skills.
The Three-Way Enterprise Agent Race
The timing of this release is not an accident. On the exact same day — 22 April 2026 — Google announced its Gemini Enterprise Agent Platform, and OpenAI's agent builder continued its own aggressive rollout. AWS, Google and OpenAI have now all converged on the same category: a managed runtime for production agents with governance, persistence and tool-use baked in.
The differentiation playbook is already visible. Google leans on its data estate (BigQuery, Workspace, Gemini's long context). OpenAI leans on the model and the developer ecosystem. AWS leans on what AWS has always leaned on: the boring, powerful parts — IAM, VPC, CloudFormation, audit logs. If your organisation already runs on AWS, AgentCore is now the path of least resistance for putting agents in front of real customers.
What This Means for Marketing Teams
Here's the uncomfortable honesty: almost nothing in this release was built for marketers. It was built for the platform engineers who support marketers.
AgentCore is magnificent infrastructure. It is also, for a CMO or an in-house marketing lead, completely the wrong layer to be working at. You don't want to pick a microVM region. You don't want to configure AG-UI. You don't want to decide whether your campaign agent needs filesystem persistence or whether a Skill bundle needs IAM scoping.
You want campaigns shipped. You want content drafted, approved and distributed. You want SEO monitored, paid spend optimised, and the weekly report written before the meeting.
The 2026 shift isn't 'every marketer becomes an AgentCore operator.' It's 'every marketing team needs an application layer that runs on top of stacks like AgentCore' — the way Salesforce runs on top of the database rather than making salespeople write SQL.
That's the problem we built Anjin to solve.
Anjin: The Marketing Operating System That Sits Above AgentCore
If AgentCore is the engine AWS built for agents, Anjin is the Marketing Operating System marketers actually use.
Anjin is a single platform that manages marketing end-to-end — content generation, campaign planning, channel distribution, performance tracking, SEO, affiliate pipelines, brand consistency — all running inside one operating system powered by agents that understand your brand as well as you do. Underneath, we handle the infrastructure choices: model routing, tool orchestration, session state, evaluation, governance. You never see a microVM.
What Anjin replaces:
- The content agency drafting and publishing across channels
- The SEO consultant optimising and tracking rankings continuously
- The paid media planner briefing, testing, reporting
- The distribution workflow — the spreadsheets, Slack threads and Notion pages holding marketing together
- The £8–15k/month you're spending to coordinate it all
What Anjin does that none of them can:
- Runs 24/7. Your agency doesn't.
- Learns your brand voice in hours, not months.
- Ships the same day a news moment breaks — AgentCore's release is a perfect example. A competitor AgentCore explainer was live on Anjin within an afternoon.
This is the category we're building: a Marketing OS that lets one operator run marketing at the scale that used to require a team of twelve. The plumbing underneath — AgentCore, Bedrock, whichever model wins this quarter — is our problem, not yours.
The £888 Lifetime License — Offer Closing Soon
Lifetime access to Anjin for a one-time payment of £888. Not a subscription. Not a seat. Not a trial. One payment, unlimited use, for as long as Anjin exists.
The average marketing team spends £888 in about three working days on tooling, freelancers and coordination software. You're buying the platform that replaces most of it — once.
This price will not be offered again once we close our early-access cohort.
Claim your £888 Anjin lifetime license →Founders, agency owners and in-house marketers — this is how you run marketing at AI speed without the team, the burn, or another year of waiting.
Sources: AWS What's New, AWS Machine Learning Blog, AWS News Blog, AWS AgentCore product page, AgentCore Browser release notes, SiliconANGLE, The Letter Two.
