deep dive on agentic AI

Artera CEO Guillaume de Zwirek Goes Deep on Agentic AI, Legacy Tech, and What’s Next

Artera co-founder and CEO Guillaume de Zwirek recently sat down with Healthcare AI Guy for a deep dive on agentic AI, how the company is rebuilding itself in this new era and what’s coming next. The conversation covers a lot of ground: why Guillaume thinks much of Artera’s own existing technology is headed for retirement, why he’s skeptical that any vertical AI vendor can out-build the frontier labs, and what it actually takes to test an autonomous agent before it touches a real patient.

A few of the specific talking points from their conversation:

  • Why Guillaume believes older technology should head for retirement — including products Artera spent years building
  • Why domain knowledge and distribution matter more than whose model is technically best
  • How a 12-clinic pediatric organization used agents to book more than 10,000 well-child visits in four months
  • Why Artera builds a synthetic replica of each customer’s actual EHR environment before any agent goes live

You can read the full conversation on Healthcare AI Guy. Below, we pulled seven moments from the interview worth watching on their own.

A lot of our older technology should eventually retire.

Hundreds of Artera customers still run on deterministic workflows, and the company built copilots that suggest messages for staff to approve before they go out. Guillaume doesn’t treat either as sacred. “Frankly, I think a lot of our older technology should eventually die,” he said. “Hundreds of customers still use deterministic flows, but rigid decision trees are a relic of the past.” This is a clear signal of how fast Artera thinks the ground is moving underneath healthcare software.

If the agent can resolve it, let it resolve it.

That same logic shapes how Artera thinks about human review. For years, the default was a copilot drafting a message and a staff member approving it before it reached a patient. Guillaume now questions that step itself: “If an agent can safely resolve the request itself, why have a person approve the text?” Scheduling, refill requests, billing questions, and insurance updates are moving toward full autonomy — the line isn’t about task type, it’s about acuity. Anything complicated, acute, or outside policy still goes to a person.

Automating the busywork between patients and care.

Asked to explain what Artera actually does, Guillaume kept it plain: “We automate the routine, undifferentiated work between patients and healthcare staff that gets in the way of people receiving care.” That includes the phone calls, texts, scheduling, eligibility checks, billing, and referrals that happen outside the exam room: work that eats staff time without improving anyone’s health. He later put a number on the ambition: Artera’s rough estimate is that the industry will “need to automate around 80% of today’s undifferentiated administrative work” just to free up enough capacity for the harder conversations that actually require a person.

Technical supremacy isn’t the moat — domain and distribution are.

With AI-native startups entering healthcare from every direction, Guillaume pushed back on the idea that Artera’s edge comes from having a better model. “I’m skeptical of a vertical AI vendor claiming it can out-innovate the frontier labs on fundamental model technology,” he said. What he thinks actually compounds over time is harder to copy: a decade of healthcare’s endless exceptions — state-by-state rules on controlled substances, patient proxies, provider-specific scheduling quirks — and relationships with more than 1,000 healthcare practices. “When we ask customers what they value, the word we hear most often is ‘partner.'”

The patient experience problem is a business model problem.

Guillaume traced healthcare’s access problems back to incentives rather than technology. “I think it’s a business model problem,” he said, pointing to a payer-reimbursement structure where patients often have few alternatives, which means friction gets tolerated instead of fixed. He made the point personal: despite having a physician for a mother, a history of heart-related issues, and running a healthcare company himself, “I haven’t seen a primary care physician in two years,” he admitted, simply because he didn’t want to navigate hold times and a forgotten portal password.

Put your builders close to the customer problem.

Artera’s AI Service Squads put engineers directly inside a practice’s workflow instead of routing every request through a traditional product pipeline. Guillaume’s example: ten customers on the same EHR can have ten different FHIR configurations, ten different scheduling exceptions, and front-desk knowledge that’s never been written down anywhere. “We put builders close enough to the problem to solve it directly,” he said, fast enough to actually keep pace with how different healthcare operations really are, rather than how they look on paper.

Test AI against a synthetic EHR and real patient personas.

Before any agent reaches a real patient, Artera builds a synthetic replica of that specific customer’s EHR environment – not a generic sandbox, but one reflecting their actual scheduling patterns, API quirks, and integration configuration. From there, the agent is tested against synthetic personas built from de-identified real interactions and scored on reliability. If a change introduces hallucinations or causes the agent to call the wrong tool, the release gets blocked before it ever reaches live traffic.

Read the full interview on Healthcare AI Guy, including Guillaume’s origin story, Artera’s testing and security architecture, and where he sees patient access heading over the next three to five years.

Related Posts

In August, an AI squad submitted a new voice-agent feature for adversarial review. The seven-persona pass came back with nothing...
Artera prepared this comparison and is one of the vendors ranked. Vendor information reflects public sources as of October 2026;...
Artera prepared this comparison and is one of the vendors ranked. Vendor information reflects public sources as of October 2026;...
Connect with Us