Mar 15, 2026
By Shuqi Mao, Head of Customer Success, Peer AI
Across biopharma, there is no shortage of interest in AI. Most organizations I speak with have already funded pilots, pulled together internal working groups, and talked openly about how automation is expected to change the way work gets done. Where things start to falter is later, after the pilot phase, when the expectation quietly shifts from experimentation to daily use. That is usually the point where progress slows and the technology stops moving forward inside real workflows.
Ultimately, when thinking about AI adoption, I return to the core of what defines humanity: our unique ability to cooperate flexibly in large groups, driven by belief in shared myths and stories. This capacity for shared belief creates a crucial layer of trust, enabling us to build complex societies and systems. Our evolution was never just about intelligence; it was about the collective trust and community we forged.
As AI now enters our most regulated workflows, we face a new inflection point in collaboration. We are no longer simply asking individuals to adopt a new tool. We are asking them to extend that foundational human trust to a non-human collaborator. In this context, the technology must overcome the highest hurdle of all: earning trust, the most valuable human asset.
Challenges in AI Initiatives
Limits of traditional change management
Many organizations approach AI adoption the way they handled earlier digital transformations, with executive sponsorship, rollout plans, communications, and formal training. For conventional software, that was often enough because usage could be mandated and people gradually adapted. AI changes the equation. Users interact with it directly, and its outputs are not fully deterministic, which shifts the evaluation from usability to trust. In regulated environments, where individuals are accountable for what leaves the organization, AI hallucinations draw intense scrutiny. What might be labeled resistance is often risk management, as teams probe the system and hold it to a higher standard before integrating it into their workflows.
When AI creates more work instead of less
Another recurring friction point is cognitive overload. In medical writing, long documents and layered approvals mean that reviewing AI-generated drafts can require substantial verification and correction. If the cost of review exceeds the cost of drafting from scratch, the perceived benefit disappears. Trust weakens, and usage slows, not because teams oppose automation, but because they are protecting their time and professional credibility. Quality becomes decisive early on. Before meaningful time savings appear, users need confidence that AI agents understand their context and produce drafts worth engaging with.
How usage grows in practice
Even when these challenges are addressed, adoption rarely spreads through mandate alone. Sustained usage tends to emerge when individuals see clear value in their own work and begin sharing that experience informally with colleagues. Trust builds first within small groups and extends outward as others observe consistent results. Early experiences carry disproportionate weight, which means unreliable or misaligned output can stall momentum quickly, while high-quality drafts help usage expand more naturally over time.
The early milestone is when users begin to feel that working with the system is easier than working around it. That shift does not come from technical demonstrations alone. It comes from consistent, context-aware output that earns trust. Once that happens, usage begins to sustain itself.
Our product development approach is shaped directly by these adoption challenges. Rather than assuming users will adapt to the technology, we try to reduce the friction of adoption from the start. That includes building systems that do not require writers to master prompt engineering, allowing AI-assisted drafting to be applied consistently across different document types and use cases.
It also means designing clear human control points into the workflow. In regulated environments, users need confidence that they remain accountable for the final output. Guardrails around hallucinations, bias, and review processes help ensure individuals stay in control while still benefiting from the automation.
Finally, we try to minimize disruption by meeting users inside the tools they already rely on. Integrations such as a Word add-in allow AI to appear within familiar workflows. This lowers the barrier to experimentation and makes it easier for teams to incorporate the technology into everyday work.
Trust as the first milestone
Expectations for immediate productivity are often high at the start. In practice, early phases are dominated by learning, iteration, and adjustment as the system is aligned to specific workflows and standards. Clear expectation-setting matters because teams are more willing to invest time when they understand the goal is alignment and quality, not instant efficiency.
The turning point tends to be consistent across organizations. Trust begins to build when users receive high-quality first drafts that fit naturally into their existing workflows. Only after that confidence is established do meaningful time savings begin to appear.
Adoption as a progression, not a rollout
Over time, it has become clear that adoption works better when it is treated as a progression rather than a single rollout. That thinking is what led us to the Crawl, Walk, Run model, which mirrors how teams actually move from experimentation to sustained use in practice.
Crawl is about getting close to how the work is actually done. That means spending time inside existing workflows, looking at past documents, and understanding how people draft, review, and collaborate day to day. It also involves working directly with customers so that this context is reflected in the system as it is being built. This approach is sometimes called forward deployment, but in practice it comes down to staying close to reality long enough for the technology to reflect it.
Walk is the phase where users start working with the system in scenarios that feel familiar. Training tends to be hands-on, with medical writers closely involved alongside engineering. Rather than introducing everything at once, this phase centers on use cases people already recognize, which helps build confidence without asking them to take on unnecessary risk early.
Run is the point where users begin to feel real ownership over the workflow. They can operate independently, with a clearer sense of when and how the system fits into their work. Support remains available, but the dynamic changes over time, shifting from close guidance to an ongoing partnership as needs continue to evolve.
In practice, organizations are often running in one area while still crawling in another. Different document types, teams, or workflows tend to move at different speeds. When adoption is treated as uniform across the organization, that mismatch can create frustration and slow progress unnecessarily.
What makes AI usable at scale
When AI succeeds in biopharma, it usually has less to do with the model itself and more to do with how adoption is handled inside the organization. Teams that invest in workflow redesign, expectation-setting, and close collaboration with users are more likely to see sustained usage than those focused primarily on incremental technical gains. In regulated environments, value emerges when trust forms and judgment is supported rather than replaced. The human layer around the technology ultimately determines whether AI becomes part of daily work or remains a pilot.
Key takeaways
AI adoption tends to stall when pilots move into regulated, accountability-heavy workflows without rethinking how work is actually done.
In medical writing, the cost of reviewing AI output can outweigh the benefit if quality is not high enough from the start.
Viewing adoption as a "progression" (like Crawl, Walk, Run) is more effective than a rigid, top-down rollout. Success is driven by deep workflow alignment and collaboration with users, not just technical depth.
Sustained usage should not be a mandate. It emerges informally when individuals see clear, personal value and begin sharing their positive experience. Trust is the first milestone, building locally and extending outward as colleagues observe consistent, high-quality, and context-aware results.
