Agentic AI in life sciences
C · Weak signal A Capgemini point-of-view paper that markets its own services; its one original figure (a scoring of 270 processes) comes without a published method, and most other numbers are recycled from MIT, McKinsey and Capgemini surveys.
Summary based on the full text·Original: about 15 minutes to read
In short
The first paper (May 2026) in a Capgemini series on "rearchitecting life sciences for agentic AI", by Thorsten Rall, Parvin Moyassari and Neeraj Goel Mittal with a long list of Capgemini contributors. It argues that pharma and MedTech AI spending rises while value stalls, and that the fix is redesigning end-to-end workflows on a shared data and governance layer instead of funding isolated use cases. Its own analysis rates about 84% of 270 life sciences processes as automatable or augmentable by agents. The paper is a positioning document rather than a study.
Key numbers
- ~84%share of 270 pharma and MedTech processes Capgemini rates as 'agentifiable'
- 8%life sciences firms actively implementing an enterprise AI strategy, against 71% defining one
- 63%biopharma R&D leaders who lack the data maturity to realise AI's impact on R&D
As reported by the source.
English·PDF·opens capgemini.com
What it reports
- From a Capgemini Research Institute survey of biopharma R&D leaders: 86% expect AI to reshape R&D but 63% lack the data maturity; 76% anticipate new ways of working but fewer than half feel ready; 71% are defining enterprise AI strategies, 8% are implementing them.
- Cites the 2025 MIT "GenAI Divide" report: $35 to $40 billion invested in internal AI initiatives, 95% of organisations seeing near-zero returns.
- Industry context: $200 to $350 billion in annual pharma revenue exposed to expiring exclusivities; cost per molecule above $2 billion; MedTech development cycles of three to seven years.
- Capgemini's process matrix: about 28% of workflows "start here" (low complexity, high impact), about 26% strategic, about 17% opportunistic, about 12% deprioritised, about 16% not agentifiable.
- Not agentifiable examples: first-in-human trial design, Qualified Person batch sign-off, on-site GxP audits, FDA advisory presentations.
- Quotes McKinsey's estimate that 75-85% of life sciences workflows contain agent-ready tasks, freeing 25-40% of workloads.
- Names five capabilities: workflow redesign, AI-ready data and semantic layer, agent orchestration, a governance control panel, and "human-AI chemistry".
- Expects headcount to fall gradually in specific functions as agents absorb junior-to-mid-level cognitive tasks, with new roles such as agent orchestrators and clinical AI validators.
How they got it
Capgemini scored 270 pharma and MedTech processes on business value, cross-functional collaboration, technical complexity and effort, and plotted them on an impact-complexity grid. No scoring rubric, rater details or process list are published. Survey figures come from separate Capgemini Research Institute reports.
Limits
- The firm sells the transformation services it recommends; no client results or measured outcomes are reported.
- The MIT 95% figure is widely disputed and is cited without caveat.
- Use cases (pre-call briefing agents, CAPA agents) are described as possibilities, not deployments with numbers.
Why it matters for companies
Read it as a checklist for regulated industries rather than evidence. Its useful parts are the list of decisions that stay human (sign-offs, audits, first-in-human design) and the advice to build only the parts of the data layer that a chosen high-value workflow needs, instead of a "fix all data first" programme. The plan to tell staff early about gradual headcount effects in junior-to-mid roles is a point HR and works councils will need to address in any regulated firm.
Summary based on the full text. Read the original for the full argument and data.Open PDF on capgemini.com
- Tags
- AI agents, Operations