News
Healthcare IT News
FeaturedWhy Johns Hopkins is benchmarking AI agents before deployment
Health system leaders say reliable benchmarking, governance and workflow design must come before scaling agentic AI across administrative operations. Johns Hopkins Medicine is taking a deliberately cautious approach to agentic AI, focusing first on proving reliability and governance before expecting measurable financial returns.
Blog
Going Once, Going Twice: Notes From the Floor of the AI Model Auction
On September 25, our favorite value model got delisted. Kimi K2.6 was cheap and good at extraction, and then its host retired it. That capped a summer when Kimi K3, GPT-6 Sol and Luna, and Claude Opus 5.5 all repriced the market. Joon Lee, Head of Forward Deployed Engineering, explains how his team picks models for healthcare customers: treat it as a reverse auction with a reserve price on quality. He walks through a real enrollment agent where 4 models bid, none cleared the bar, and redesigning the job changed the outcome.
Video
Actava.ai | Agentic Workforce | Agents That People Actually Adopt
The most common ending for a working agent is not failure. It is quiet non-use. Why good agents go unused is predictable: nobody asked, the agent is not where the work is, or the win is invisible. Adoption starts with a problem the team already feels, inside the tools they already use, with value they can see. Ask the team what they hate. Build that. It gets used.
Resource
ACTAVA | Payer Operations | Claims Review Agent
A generic Payer scenario. Closing a claims review backlog without opening a compliance gap. Claims review has a specific failure mode. The work doesn't stop coming, but the people qualified to judge it don't scale with it. Subject-matter experts are the actual bottleneck, and every decision has to hold up on a later audit, not only clear the queue today. An agent built for this starts from the same ground truth a human reviewer would use. It uses the organization's own policy documents, uploaded once and kept current, rather than a general model's guess at what a policy probably says. Its plan is written as explicit steps, each one citing the exact document it draws from. Where a decision would change what someone is owed, that step is marked as requiring approval before it can run. A hard stop written into the plan, not a suggestion. That approval does not go to whoever happens to be online. It goes to a named group with its own policy, whether all must approve, a majority, or any one qualified reviewer, and a timeout if nobody responds in time. While it runs, the agent searches the knowledge base live and shows exactly which passages it found and cited. A reviewer checks a specific citation rather than taking the agent's word for it. And we benchmark its accuracy against a real dataset, run after run. Whether it is working becomes a tracked score over time rather than an impression. Grounded in your own policy. Gated by your own people. Measured the same way every time. That is a review agent built to be trusted, not only fast.
Release Notes
ACTAVA Platform Release Notes CHRYSO v8 (September 2026)
ACTAVA now tracks AI compliance where the AI runs. CHRYSO, shown in the platform as Compliance, measures an organization against frameworks such as the NIST AI Risk Management Framework and HIPAA technical safeguards, and recalculates every control from what is happening on the platform: live agents, their evaluations and run history, approved policies, signed acknowledgments, completed training, and uploaded evidence. Organization administrators get a single page showing how many controls are satisfied, what is blocking the rest, and who needs to act. Members and citizen developers get a short task list of policies to sign and training to complete. For healthcare payers, providers, and life-science teams, the result is an audit-ready posture that stays current without a separate compliance spreadsheet.
Release Notes
ACTAVA Platform Release Notes KORA v8 (September 2026)
Actava.ai KORA v8 opens the platform to teams building on it. One organization API key now creates, versions, runs, schedules, benchmarks, and exports an agent, and mandatory turn gates hold a draft answer until its required checks run. Pipeline Builder and Voice Call Replay arrive in beta.
Video
Actava.ai | Agentic Workforce | Agents That Someone Is On Call For
The first agent to fail at three in the morning will reveal whether autonomy was a decision or an assumption. An unattended agent still needs an operating model: a signal that fires on failure, a runbook that says whether to retry, disable or escalate, and an off switch that does not depend on the person who built it. Autonomy needs a pager, not just a dashboard. Decide who gets woken up. That decision is part of the design.
Video
ACTAVA | Release | CHRYSO v8 (September 2026)
ACTAVA now tracks AI compliance where the AI runs. CHRYSO, shown in the platform as Compliance, measures an organization against frameworks such as the NIST AI Risk Management Framework and HIPAA technical safeguards, and recalculates every control from what is happening on the platform: live agents, their evaluations and run history, approved policies, signed acknowledgments, completed training, and uploaded evidence. Organization administrators get a single page showing how many controls are satisfied, what is blocking the rest, and who needs to act. Members and citizen developers get a short task list of policies to sign and training to complete. For healthcare payers, providers, and life-science teams, the result is an audit-ready posture that stays current without a separate compliance spreadsheet.
Video
Actava.ai | Agentic Workforce | Agents You Shouldn't Build
The fastest way to lose credibility with an operations team is to bring an agent to a problem that wanted a checkbox. Agents are good at judgment under ambiguity. If the rule is deterministic, write the if-statement. If the input is structured, use a query. If a single mistake would be unacceptable and unreviewable, add a person—or do not automate. Knowing where an agent does not belong is what makes the rest of your agentic workforce credible.
Blog
Inside Agent Workspaces: How Actava.ai agents hand off work without handing off permissions.
Packet 37 in a 200-packet claims batch is missing its itemized bill. A document-review agent finds the gap, a claims-review agent carries it forward, and a communications agent needs a human's approval before it asks for the bill. Frank Wang follows that one packet through Agent Workspaces in KORA to show how work moves between specialists while each one runs on its own permissions, how an outbound request pauses for approval, and how a Release Barrier blocks an assistant from saying "packet complete" when the evidence still says otherwise. He closes with the trade-offs the team accepted and five design choices you can borrow for your own agent stack.
Video
Actava.ai | Agentic Workforce | Agents That Start With One Boring Workflow
Ambitious first agents collapse under their own scope. This video argues the opposite: pick a frequent, rules-based job with a clear input, output, owner, and exception path; run it alongside the manual process; and widen autonomy only after you've seen where it fails. The fastest way to overcomplicate an agent is to start with the most ambitious workflow. Begin with one boring job instead: frequent, bounded, rules-based, and easy to measure. Give it a clear input, output, owner, and exception path. Run it beside the manual process, watch where it fails, and improve it before widening autonomy. Small, observable wins build the trust and operating discipline an agentic workforce needs.
Video
Actava.ai | KORA | Agent Workspace (Detailed)
Multi-agent work fails when handoffs blur the difference between what was requested, what was handled, and what was actually completed. Actava.ai KORA Agent Workspaces keeps run-owned state moving while replacing agent-owned authority at every transition. Tools, knowledge, models, verification policies, and approvals belong to the next Agent Card—not to inherited context. Approval happens before irreversible actions. Evidence comes from runtime-issued receipts. Missing or malformed checks block release. The result is coordinated specialist work with explicit boundaries, traceable evidence, and controlled delivery.
Video
Actava.ai | Release | KORA v8 (September 2026)
V7 settled who's accountable for the agents. V8 settles who gets to build with them. Here's what shipped, what's still in beta, and what needs configuration before you roll it out. V8 ships 82 items, and one thread runs through all of them: other teams can now build on the platform. A partner can create, version, run, schedule, benchmark, and export an agent with an organization API key, without opening the dashboard. A team can require checks before an answer goes out. A downstream system can receive results in a schema it declared up front. Actava.ai released v8 with 82 product improvements, all around one idea: openness. Partners can work directly in shared delivery surfaces, teams can trace and check the evidence behind an answer, and workflows can run from intake through completion without hiding the handoffs in between. This overview covers the partner API, mandatory turn gates, structured results, Pipeline Builder, voice replay, new connectors, compliance controls, and the operational details teams should review before rollout. The result is a more open operating model for enterprise agents: visible work, clearer accountability, and stronger control across the full lifecycle. V8 is a release for teams embedding governed agents in their own products. ACTAVA KORA covers Build, Test, and Learn, and ACTAVA Compliance covers Guide. V8 extends both to the people who build on top of them.
Showing 1–13 of 248 results
From production workflows to customer-controlled intelligence.
Don't give your agentic future away to a single model provider. Don't mistake consumer tools for real, safe Enterprise Agentic Tools. Enable your citizen developers to create and manage the AI Agents they need to run their part of your business.
Build complex agents. Test their reliability. Learn from every workflow. Own your intelligence.