Sign in

News & Blog

Research, engineering, and ideas from Actava.ai.

Videos

117 videos

VideosWatch video

ACTAVA | Agentic Workforce | Agent KPIs

How ROI measurement works in Actava.ai, from org-level value drivers through each agent's justification packet to live projected-versus-actual reporting. Here is what we see across healthcare. AI is everywhere. Returns are not. Most teams can tell you whether an agent ran and what it cost. Very few can put a defensible number on the outcome that agent was hired to deliver. Actava.ai closes that gap by treating every agent as a Non-Human Resource. A digital employee. It earns a salary, which is its run cost. It works overtime. It carries a job description. And it gets a performance review every single day. Measurement starts at the organization level in ROI and Value Settings. Value Drivers are the business outcomes you measure against, ranked Gold, Silver, and Bronze, each with a target and a dollar-impact hint. Drivers become the shared vocabulary for every agent you build, so each agent's KPIs map back to what the organization actually cares about. Alongside them sits your Agent Building Cost Model: a base cost to stand up any agent, a cost per capability step, and annual maintenance as a share of build. Every run is metered in Orchestration Units, which keeps run costs comparable across agents and orgs and lets finance budget for your Total Cost of Agent Capacity, or TCAC. At the agent level, every agent carries a packet called Why We're Building This, computed from your org settings. It states build cost, maintenance, a viability score, the plain-language business goal, the challenge addressed, and the limitations, including where a human stays in the loop. The packet locks when the developer submits the agent for approval. One detail matters more than any other: if none of an agent's KPIs are bound to a corporate value driver, Reportable ROI reads "not measurable." The platform will not report ROI it cannot actually track. Bind a KPI to a driver, point it at a measurable data source, and the number becomes real. Once an agent goes live, projections turn into measurements. Real Orchestration Unit spend replaces the estimate, KPIs record actuals against baseline, and improvement converts straight to money where a driver carries a dollar hint. Your CFO sees both sides of the ledger in one panel: projected, which answers "was this worth building," and actual, which answers "is it delivering." Daily health scores roll it up into one recommendation per agent. Scale it, fix it, or stop it. Your human workforce justifies its headcount every year. Your digital workforce should too.

Watch ↗

VideosWatch video

ACTAVA | Agentic Workforce | Building Your Agentic Workforce

Once your roadmap is set, this is how Actava.ai turns a prioritized workflow into a governed, production-ready agent. An AI agent is an autonomous digital worker that perceives context, plans and executes multi-step work through the systems you grant it access to, and operates within guardrails you define. It should be supervised for outcomes and measured for ROI the same way you would manage a human team. Building one starts the way hiring does. You would never hire a "Generic Worker." You hire a Referral Manager or a Compliance Officer. So instead of a generic prompt, you define a specialized persona with a targeted outcome, and the agent plans its own steps, executes, and self-corrects. Actava.ai Agent Studio makes that buildable by people who know the work. The Agent Builder Wizard starts a dialogue rather than presenting a blank form: describe the role as you would to a human recruit, and AVA generates a sophisticated first draft of the identity and operational plan directly into the builder. From there, you configure what makes the agent real. Grounding through a scoped knowledge base, structured at the organization, project, or agent level. Memory, which captures corrections and preferences learned over time as a layer on top of a stable system prompt. Data access through Model Context Protocol connectors to EHRs, payer portals, claims systems, and the rest of the archipelago healthcare runs on. Tools and skills: a tool is what an agent has, and a skill is packaged know-how for how it does the work. And human-in-the-loop approval, set where the stakes require it and relaxed as the agent earns trust. Guardrails are configured alongside capability, not after it. Every employee needs to know their no-go zones, and in healthcare and life sciences that is a legal requirement, not a best practice. Agents get the same treatment: hybrid retrieval that grounds every response in your data with full citation provenance and audit trails, plus automatic PHI detection and redaction at retrieval time so sensitive data stays protected while the agent works. The result is a testable agent in days rather than quarters, built by the people closest to the workflow, with compliance posture and operating cost decided deliberately rather than discovered later.

Watch ↗

VideosWatch video

ACTAVA | Agentic Workforce | Planning Your Agentic Workforce

How to decide where agents should actually operate in your organization, and in what order, before you build a single one. McKinsey estimates 70 to 80 percent of enterprise work can be agentified. The value is real, pervasive, and substantial. But most organizations fall into a common trap: they jump straight to deploying agents before understanding where those agents belong. They automate whatever is top of mind, or whatever the loudest voice complains about. The result is a scattered collection of automations that never add up to real progress. Before you prioritize, you have to see the full picture. Most organizations lack a systematic inventory of their operational workflows, including how often they run, who handles them, and the friction they create. That inventory is the starting point. Agent-ready workflows tend to fall into a few clear patterns: high-volume, repetitive processing; cross-system data movement where humans serve as the glue between platforms; information gathering and synthesis; triage and routing; and monitoring with exception handling. From there, Actava.ai assesses workflows on two dimensions. Automation potential asks how feasible the work is: predictable sequences, structured inputs, delegable system access, measurable success, and few irreversible judgment calls. Operational impact asks how much it matters: hours consumed, frequency, error cost, and human toll. Plotting the two produces four waves. Quick Wins are high potential and high impact, and they go first to build momentum. Strategic Bets are lower potential but high impact, and they justify custom agents or hybrid human-AI approaches. Low-Hanging Fruit builds team skill without leading the program. Everything else defers. Four sequencing principles keep the roadmap honest. Begin where the pain is loudest, because buy-in drops fast if early projects miss what people actually care about. Build in public with clear before-and-after metrics. Respect dependencies, since some large workflows only become feasible after lower-level processes are standardized. And measure human impact beyond efficiency, because reduced cognitive overload and lower stress are often the strongest case for continued investment.

Watch ↗

VideosWatch video

ACTAVA | Agentic Workforce | The Agentic Workforce Playbook

The Actava.ai organizational playbook for implementing agentic AI: how executive teams move from experimenting with AI to running governed agent workforces that perform real work, safely and at scale. Most organizations have spent the past few years getting comfortable with generative AI as a personal task assistant. You prompt it, it responds. The next shift is bigger. We are at the beginning of the era of hiring non-human resources to work alongside your teams, partners, and customers, reasoning on your behalf toward outcomes you set. Agentic systems pursue objectives. They plan, call tools, hold context across steps, and self-correct as they execute multi-step work inside your business processes. That changes the stakes. An assistant's mistake stays inside a conversation. An agent's mistake propagates into decisions and operations. This video walks through the six things leadership teams need to get right. Align your agentic vocabulary before your strategy, because the most common failure point is not technical: each leader at the table quietly defines "agent" differently. Design around workflows, not tools, with clear inputs, outputs, and success criteria. Calibrate autonomy to the stakes rather than defaulting to the maximum, starting with tight human-in-the-loop oversight and widening as trust and measured performance grow. Hire and manage your agents rather than only building them, through the AgentOps discipline of continuous monitoring, drift detection, and outcome evaluation. Scale through governance and safeguards: hardcoded constraints, output verification, reversibility, and escalation paths to human judgment. And measure the outcomes and cost of every agentic workflow, anchored in Total Cost of Agent Capacity (TCAC). Success with agentic AI depends far less on the underlying model than on workflow design, governance, safeguards, and organizational readiness. Treat it as a managed operational capability, grant autonomy as it earns trust, and it becomes a dependable operational partner rather than a risky experiment. Closing the gaps across people, process, and technology determines which enterprises thrive in the era of agentic AI. Actava.ai delivers the first agentic lifecycle suite built for healthcare, designed to close all three gaps together so your digital workforce operates seamlessly alongside your people and systems.

Watch ↗

VideosWatch video

ACTAVA | Release | KORA v5 (May 2026)

Version 5 of the Actava.ai platform shipped in May 2026. From agents that work to agents that scale — and a benchmark published against itself. 10 new workforce agents, six major platform capabilities, six integration and governance upgrades, and 31 more improvements underneath. It opens with something most vendors do not publish. χ-BENCH tests healthcare administration across policy-dense workflows, multi-role coordination, hidden state, and provider-payer handoffs. The best agent setup solved 28% of complex tasks on the first attempt. None cleared 8% under consistency testing. Nearly a third of the tasks failed for every model-agent combination tried. And on realistic provider-payer handoffs, performance fell to 0%. The conclusion is worth hearing in full. The defining question for healthcare in 2026 is not whether agentic AI will work. It is whether the industry can reliably detect when it does not. Everything else in this release is an answer to that. The ten new agents go straight at healthcare administration. Provider credentialing across the full NCQA workflow with sanction screening. Payer enrollment and network adequacy. Plan configuration with parity testing. Eligibility, onboarding, prior authorization, and continuous monitoring of the Federal Register. Then, the capability that enables them to scale. A pipeline can start from a list — a connector or an uploaded file — fan out into thousands of parallel runs with strict per-row isolation, reduce the results to a single report, and resume automatically after an infrastructure failure. Approval moved up a level. Reviewers approve the work an agent does, not the API calls it makes, routed to named groups like Clinical Reviewers or Compliance, under a policy applied the same way whether the run came from the interface or the API. Multi-organization architecture provides a single identity across tenants, with no ambient access between them. Scoping is enforced per request at the service level, extended to evaluation and synthesis jobs, and carried through to the integrations themselves. The MCP Store imports a vendor API straight from an OpenAPI spec and generates an adapter for it, running on its own service with OAuth 2.1, PKCE by default, and credentials isolated per run. The model became a setting rather than an assumption. Five providers behind one factory, selectable per agent or per turn, persisted through the lifecycle, with reasoning capture and duration breakdowns wherever the run went. Voice agents reached production. Interruption control tuned per agent, a scripted test mode with importable dialogues so you can validate before anyone real is on the line, voice as a valid pipeline step, and shared version history with build agents. And the agent-card optimizer rewrites an agent from its own observed run history, preserving the mandatory built-ins and the approval flow, with a reported reduction in token usage of up to 90%. Four things before you upgrade. Slack may need re-authorizing. The default model moved, though explicit pins are unaffected. Older model rates were removed, so an ROI dashboard anchored to one of them will show a gap. And the standalone voice-runs route is retired. Scale is the easy half. Knowing when it is not working is the rest. Full release notes: https://actava.ai/news/actava-platform-release-notes-v5-may-2026

Watch ↗

VideosWatch video

ACTAVA | Agentic Workforce | Ava Director

A team of agents, coordinated, observable, and governed on one shared thread. Ava directs the agents in your conversation. Select or @mention the agents you want, and Ava runs them for you: immediately, in the background, or scheduled for a later time in your time zone. Results come back into the thread, and when an agent already has a run open, your follow-up goes into that same session so the agent keeps its context. Governance is built into the dispatch itself. Ava dispatches only the agents you have access to, and names who can approve access when you don't. Ava doesn't guess which agent you meant: dispatch is explicit, so every run is traceable to a request, a requester, and a permission. That is what makes a multi-agent thread auditable rather than merely impressive. This is what complex multi-agent orchestration looks like when it is designed for regulated operations. Agents coordinate across systems through a standardized interoperability layer, not one-off integrations. Role-based permissions, audit trails, and human approval workflows stay attached to the work, not bolted on afterward. And because every run is metered, the cost and outcome of a team of agents stay as visible as the cost and outcome of a single one. Actava.ai KORA is the healthcare-native suite for building, governing, and improving AI agents across enterprise operations, giving healthcare organizations a single system to develop, orchestrate, evaluate, and measure agents over time

Watch ↗

VideosWatch video

ACTAVA | Release | KORA v4 (April 2026)

Version 4 of the Actava.ai platform shipped in April 2026. Text, voice, and the systems your people already live in. Four major platform capabilities, six enterprise integrations, 28 items in all — and more than 100 prebuilt agents arriving with them. Our voice agents take calls and place them; they have a real number, and they are built in the same studio as everything else: speech configuration, version management, and the same lifecycle. Every call it makes is recorded, searchable, and comparable, with outbound calls tracked in real time as they run. Opening one gives you the duration and the cost split by speech-to-text, text-to-speech, and telephony. Post-call intelligence extracts a summary, sentiment, and action items, and writes them into the workspace as artifacts that the next step can use. The second capability chains agents together. A pipeline sequences them into a single workflow within a shared workspace, executed server-side with streaming for up to 6 hours, persisted so a run can be recovered — and with event replay, so a client that joins late still sees everything that happened before it arrived. Then the systems people already work in. Microsoft 365 arrived as seven services and more than 60 tools — Outlook mail and calendar, OneDrive, SharePoint, Excel, Word and contacts — alongside Slack and Google Workspace. Which is only safe with something in front of it. Authentication is OAuth 2.0 with PKCE and automatic refresh; there are nine configurable write gates that determine what an agent may actually do, plus a domain allowlist, per-service toggles, and incremental consent. MCP servers gained OAuth too, so a third-party server can be connected and disconnected properly, with tokens refreshed rather than pasted. And the agents themselves. 13 for value-based care — chart review looking forward and backward, quality gap identification and closure, readmission risk, social determinants — and 59 for people operations across 12 domains, from recruiting through to offboarding. Underneath, the unglamorous half. Synthesis gained checkpointing, so a long run resumes rather than restarting. Datasets became reusable. Post-call extraction gained a HIPAA mode for masking. Deploys went to three a day, and the event bus moved to durable streams that can be replayed. Text, voice, and the systems your people already live in. Full release notes: https://actava.ai/news/actava-platform-release-notes-v4-april-2026

Watch ↗

VideosWatch video

ACTAVA | Release | KORA v3 (March 2026)

Version 3 of the Actava.ai platform shipped in March 2026. Full participants in enterprise workflows — communicating, costed, and visible. This is the release where agents stopped sitting beside the work and joined it. 40 items across five categories, and the notes name what that takes: professional communication, cost visibility, and a way to see what an agent actually does. Email first. An agent composes a branded message from your own templates, individually or in a batch, and the approval is part of sending it rather than something bolted on afterward. Then Slack, where approval grew up. Multi-approver workflows with a proper OAuth flow, group management, and summaries generated for the approver — so the person signing reads a decision rather than a transcript, with a confirmation step before anything commits. Then visibility. The builder draws the agent as a diagram: its steps, its tools, and the dependencies between them, detected rather than declared by hand, with an audit panel beside it. A wall of configuration is hard to check. A picture of it is not. And cost became visible. A usage dashboard breaks down spend by category, with evaluation billing and grading tokens tracked as their own line item — because the cost of measuring an agent is real, and treating it as free is how a budget goes wrong quietly. Alongside it, limits you can see coming. Organization quotas block a run past its ceiling rather than letting it through, and a context indicator shows the token budget in real time as a conversation grows. Then research. Three purpose-built search tools rather than one general one — peer-reviewed literature, repositories and code, and professional profiles — so an agent doing desk research is not reduced to a web search and hope. And what an agent produced started to look like it came from your organization. Branding is applied across generated HTML artifacts, and a map tool renders interactive maps with markers and descriptions: a facility locator or a provider directory as an output in its own right. The studio took 14 changes of its own — a guided builder wizard, plans structured into sections, complexity modes carrying their own model overrides, and a run history you can actually search. Underneath, the evaluation judge moved to Opus, access controls became organization-scoped, and steering a run mid-flight now cancels the tool call it interrupts rather than racing it. Communicating, costed, and visible. That is a participant. Full release notes: https://actava.ai/news/actava-platform-release-notes-v3-march-2026

Watch ↗

VideosWatch video

ACTAVA | Release | KORA v2 (February 2026)

ACTAVA's March 2026 release puts agents inside the tools your teams already use The March release connects ACTAVA agents to Google Workspace and Slack, gives administrators a clearer view of what agents cost, and tightens reliability for demos and production. Agents work where clinicians already work. Agents now operate across Gmail, Google Calendar, Google Drive, and Google Slides through 40 new tools. They can run referral follow-ups, schedule appointments, and generate care-plan documents. Administrators switch individual capabilities on or off per agent, and the team removed Drive delete operations from agent tooling to keep production safe. Humans stay in the loop through Slack. A full Slack integration lets agents pause a workflow and ask a designated channel for approval. Every notification is tracked, which gives compliance teams an audit trail.

Watch ↗

See what ACTAVA can do for you

Bring one workflow your team still runs by hand. We'll show you the agent that runs it.