Uploaded October 2026 | Updated November 2026, 1 day ago
How do you bring AI agents into pharma workflows without losing control? A talk from the Validated AI for Pharma Summit.
Speaker:
Marcin Dubel, Appsilon (Head of Technology)
Marcin starts from a simple idea: pharma already has strong processes for validating human work, and an AI agent is a new team member that needs the same rigor. Adoption isn’t about handing people a copilot subscription. AI has to be part of the process, observable, and owned by someone accountable. He proposes a checklist of nine properties every agentic workflow system should have, grouped into ownership, evidence and leverage, and he uses Mediforce, the open-source tool Appsilon built around them, as the example.
What he talks about:
- Ownership: open source your quality team can inspect, no lock-in to one model provider or tool, and running on your own infrastructure so the data stays with you
- Evidence: reviewing every run step by step, reproducing the environment of each agentic step (prompt, tools, Docker image), version control for every change, and a named human accountable for each workflow and step
- Leverage: sharing and reusing skills across teams, humans who lead the process instead of blocking it, and keeping the cost of running AI under control
- Q&A on how long it takes to go from concept to validated AI in production, and how to tell when a task is ready to move to an agent
Marcin's main advice: take the checklist to your team, decide which properties are must-haves for you, and move step by step. Validated AI won’t happen overnight.
Recorded on 24 September 2026 at the Validated AI for Pharma Summit, hosted by Appsilon.
All summit sessions on demand: events.ringcentral.com/events/validated-ai-for-pharma-summit?utm_source=youtube&utm_medium=video&utm_campaign=validated_ai_summit_2026&utm_content=ai_adoption_overnight
More about the summit: appsilon.com/validated-ai-for-pharma-summit
Talk to Appsilon about validated AI workflows: appsilon.com/contact-us
#ValidatedAI #AgenticAI #GxP #PharmaAI
How do you bring AI agents into pharma workflows without losing control? A talk from the Validated AI for Pharma Summit.
Speaker:
Marcin Dubel, Appsilon (Head of Technology)
Marcin starts from a simple idea: pharma already has strong processes for validating human work, and an AI agent is a new team member that needs the same rigor. Adoption isn’t about handing people a copilot subscription. AI has to be part of the process, observable, and owned by someone accountable. He proposes a checklist of nine properties every agentic workflow system should have, grouped into ownership, evidence and leverage, and he uses Mediforce, the open-source tool Appsilon built around them, as the example.
What he talks about:
- Ownership: open source your quality team can inspect, no lock-in to one model provider or tool, and running on your own infrastructure so the data stays with you
- Evidence: reviewing every run step by step, reproducing the environment of each agentic step (prompt, tools, Docker image), version control for every change, and a named human accountable for each workflow and step
- Leverage: sharing and reusing skills across teams, humans who lead the process instead of blocking it, and keeping the cost of running AI under control
- Q&A on how long it takes to go from concept to validated AI in production, and how to tell when a task is ready to move to an agent
Marcin's main advice: take the checklist to your team, decide which properties are must-haves for you, and move step by step. Validated AI won’t happen overnight.
Recorded on 24 September 2026 at the Validated AI for Pharma Summit, hosted by Appsilon.
All summit sessions on demand: events.ringcentral.com/events/validated-ai-for-pharma-summit?utm_source=youtube&utm_medium=video&utm_campaign=validated_ai_summit_2026&utm_content=ai_adoption_overnight
More about the summit: appsilon.com/validated-ai-for-pharma-summit
Talk to Appsilon about validated AI workflows: appsilon.com/contact-us
#ValidatedAI #AgenticAI #GxP #PharmaAI










