We didn’t deploy AI to do our jobs. We built it into how we work.

Posted in Uncategorized

Twelve months ago our team integrated Claude Code into how we operate. We expected faster development cycles. What emerged was something structurally different: an AI layer woven into the operational fabric of the business, not layered on top of it.

That distinction matters. And it’s worth being precise about why.

The actual problem: operational memory, not speed

Building healthcare technology as a lean team means carrying context across the whole stack: product, infrastructure, compliance, data architecture, simultaneously. The bottleneck isn’t writing code. It’s knowing what the code is supposed to do, why it was built a certain way, and what will break if you touch it.

That context decays. With every context switch, something is lost. Status emails and documentation slow you down without fully solving the problem.

Claude Code doesn’t solve the speed problem. It solves the context problem. That’s a completely different thing.

What the team’s workflow actually looks like

On a typical working day, the AI layer handles the operational groundwork that used to fracture the team’s attention:

• Running QA against a spec and flagging where the output diverged, before it reaches review
• Monitoring infrastructure services and surfacing anything that’s drifted
• Catching a data pipeline anomaly before it corrupts a proof report
• Drafting communications, then checking its own output against our content rules
• Surfacing analytics from real user behaviour in a readable summary for the team to act on

Each of these requires holding the right context. The AI layer does that. The team makes the calls.

Why the gates matter more than the capability

In a regulated industry, the question isn’t “what can the AI do?” It’s “what decisions can the AI make without human sign-off?”

In our business, the answer is: fewer than you might expect.

We have hard gates in place. External content requires human review before publication. Anything touching client data requires explicit authorisation from the team. Destructive operations require a second confirmation. Those aren’t limitations on Claude Code; they’re the right design for a healthcare software product where the evidence layer is the product.

The value isn’t that the AI acts autonomously. The value is that it does the right preparatory work so that when a decision reaches a person, they have everything they need to make it quickly and well.

What this looks like in medical education specifically

Our platform captures every interaction HCPs have with educational content: question responses, confidence ratings, behavioural indicators across a programme. The end deliverable is a proof document: evidence of distance travelled, behaviour changed, not just content consumed.

That proof layer is what makes the product commercially defensible. Pharma procurement has become sophisticated. Attendance certificates aren’t enough. They want a measurement infrastructure.

Building and maintaining that infrastructure as a lean team, across an active client base, with a continuously evolving codebase, that’s where operational context becomes the differentiator. The AI layer holds that context. The team decides what to do with it.

The honest reflection after twelve months

Three things stood out:

Operational continuity is the real advantage. The speed benefit of AI-augmented development is real but modest. The continuity benefit is significant. There’s no context loss between sessions, no handover problem, no “remind me where we were.” The system picks up where the team left off.

QA changed from periodic to continuous. The old pattern: write, push, find the edge case later. The new pattern: write, use Claude to surface the edge case, fix it before pushing. The bugs that used to survive to production rarely do now.

Trust is calibrated, not assumed. Twelve months in, the team knows exactly what the AI layer handles without review, what we want to see before it acts, and what stays firmly in human hands. That calibration is the work, and it’s worth doing carefully in any regulated context.

What this means for teams in regulated sectors

If you’re building in healthcare, pharma, or any regulated environment and wondering whether AI has a role in how your team operates, the answer is yes, but the design question matters more than the capability question.

Autonomy without structure isn’t a feature. An AI operational layer with the right oversight built in is a structural capability advantage, one that compounds as the system matures.

The teams who figure out that distinction, not just which tools to use, but how to structure them into the way the business works, are building something materially more robust. That’s what twelve months with Claude Code taught us.

Related Reading

• Distance Travelled in Medical Education: What MedComms Agencies Need to Know in 2025
• Proof Reports in MedComms: What Pharma Clients Are Now Demanding
• HCP Education in Biotech: Why Measuring Behaviour Change Is the Only Metric That Matters