Atlas Agent Engine is the managed runtime that hosts MongoDB agents: it gives each one its own Atlas identity, read-only tools, and an approval gate before any write. Here is the loop, step by step.
Atlas Agent Engine hosts agents that handle routine database work, but the engine is only half the story. A run starts with a trigger, resolves a scoped Atlas identity, calls tools through the MongoDB MCP Server, and passes through guardrails that default to read-only, block outbound traffic to unallowed hosts, and forbid shell execution. Write tools pause at a human approval gate before they touch data, and memory keeps the agent from relearning the same context every session. The result is a mental model for deciding which steps belong in the managed runtime and which belong in your own code.
The MongoDB Agents templates give every agent its own Atlas identity and read-only access by default, and they require human approval for every write tool. That keeps a wrong or over-broad tool call from changing data, and it limits how much a compromised agent could do. Writes become an explicit, reviewable step instead of a side effect.
Memory is a separate layer from tool execution. Databricks reports that agents with persistent memory tripled pass rates on real enterprise tasks while using 31% fewer tokens, because they stop replaying full histories. Treat it as a cost-and-quality decision you make alongside the runtime, not something the tool-calling loop gives you for free.
The posts behind it
Databricks · · AnnouncementDatabricks announces Agent Memory on its Agent Bricks platform, now in Beta. Agent Memory improves agent quality by storing critical facts, preferences, and workflows that agents should remember, reducing both cost and latency. It is built on the MemCurrent architecture from Databricks Research, which combines context logging plugins, on-demand tool access, and a background dreaming pipeline to consolidate memories, avoiding the cost of replaying full histories or building bespoke memory systems, all served on Lakebase for low latency. In internal benchmarks, agents with memory tripled pass rates on real enterprise tasks while using 31% fewer tokens compared to stateless agents. Agent Memory joins other Agent Bricks capabilities including hosting, Sandbox, Unity Gateway, pre-built MCP servers, and tracing and evaluation, providing a complete platform for building production agents at scale.
Intercom · · AnnouncementFin 4 introduces a Customer Agent that uses Skills to autonomously resolve problems across the full customer lifecycle and an Operator agent that runs complex operations proactively through Routines and Connections. The second-generation Copilot proactively searches tools, summarizes issues, drafts answers, and recommends actions when a rep opens a conversation, making human replies faster and more informed. Human in the Loop enables humans to approve, edit, or take over conversations, shifting reps from answering to supervising Fin and designing the customer experience. Fin Apex 2.0 and Apex Flash 2.0 improve resolution rate, speed, and accuracy, while Apex for Procedures extends specialized model handling to complex multi-step queries. The helpdesk is reimagined for a world where 90–95% of customer communication will not involve a human, with Copilot and Human in the Loop working natively in Intercom and integrating with other helpdesks.
Ramp · · Problem & solutionRamp, a fintech company, transformed from a traditional SaaS provider into a fully agentic platform, building Laminar as its in-house agent harness to drive this shift. After experimenting with external frameworks such as OpenAI Agents SDK, Claude Agents SDK, LangChain, and multiple internal passion-project agent frameworks, the team found the experience uncohesive and decided to consolidate. In May 2026, they created Laminar, a common platform named to evoke smooth, structured flow, aiming to abstract away common agent infrastructure so developers could ‘build agents, not infrastructure.’ Laminar now serves as the backbone for Ramp’s agentic products, including automating month-end close for accounting firms, tracking and optimizing token spend for finance teams, and providing a personal AI assistant to cardholders via SMS. Additionally, Ramp has built an in-house collaborative cloud coding agent and a desktop agent for knowledge work, all running on Laminar, demonstrating the platform’s power for real-world product development.
Supabase · · AnnouncementTeams with production apps often learn about problems from users, then dig through logs and reports. Supabase released tools that let coding agents check production projects for error spikes, security issues, slow queries, and resource growth on a schedule. Agent prompts report each new or worsened finding with the measured value and a next step, using thresholds from public detection checks that teams can override. Other new tools include a query_logs MCP tool for read-only SQL on logs, updated agent skills that fetch current observability docs, a Health Check Advisors tab, and a Database Connections page. Agents can be run on demand or scheduled as Claude routines, Codex tasks, or Cursor automations, and clean runs report nothing.
MongoDB · · AnnouncementTeams increasingly want agents that handle routine database work—such as checking cluster alerts, reviewing slow queries, and answering questions about a schema—but building agents takes much more than a prompt, including credential handling, access scoping, tool orchestration, approval flows, and more. Today MongoDB launched MongoDB Agents, a collection of open source agent templates that combine the MongoDB MCP Server and MongoDB Agent Skills with the plumbing you would otherwise build yourself. Each template includes scoped MongoDB Atlas credentials, approval flows for writes, and a ready-to-run deployment configuration for Atlas Agent Engine. Built-in guardrails give each agent its own Atlas identity, default read-only access, require human approval for every write tool, block outbound traffic to unallowed hosts, and prevent shell execution. You can use the agents in the Agent Engine Playground, via the agentengine CLI, or integrate with Slack using the optional bot to ask questions, approve writes, and schedule reports.
Auth0 · · TutorialAI agents break three OAuth assumptions: static permissions, predictable life cycles, and direct delegation. Auth0’s Strict Mode for APIs enforces a permission ceiling by requiring explicit Client Grants and PKCE, structurally preventing rogue dynamically registered clients. Client Initiated Metadata Documents (CIMD) allow agents to host a metadata document at a stable HTTPS URL; Auth0 fetches and validates it to issue a client ID with automatic default permissions. RFC 8693 Token Exchange enables multi-hop delegation: the agent presents the user’s token and its own credentials to Auth0’s /oauth/token endpoint, which returns a token with the original user as ‘sub’ and the agent’s client ID as ‘act’. The post walks through a workflow from CIMD bootstrap to user consent to token exchange, and notes that record-level access requires Auth0 Fine-Grained Authorization.
More like this
California's SB 53 is a trust-but-verify rule for frontier AI: large developers disclose rather than follow technical mandates. This flow maps the compute threshold, safety frameworks, transparency reports, and the 15-day incident clock.
An air gap is a one-way artifact pipeline, not just a blocked network. This explainer follows six stages, from mirroring dependencies outside to operating inside with no egress.
Apache DataFusion is an embeddable query engine: SQL goes in, runs as columnar Arrow batches inside your own process, and results come out without a separate service.
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
One useful email. Unsubscribe anytime.
