Teleflex [NYSE: TFX]

AI isn’t magic. It won’t turn mediocre processes into great ones and it won’t obsolete human expertise.

It can relieve bottlenecks that slow many device programs due to the scarcity of timely, high-quality expertise. AI can reduce queue time, tighten feedback and decision loops and help teams resolve problems sooner. Used poorly, it can also do the opposite: create more confusion, more false confidence, and more rework.

Much of our cycle time is work waiting, waiting for a reviewer, waiting for clarification after a handoff, waiting for a new meeting to decide because the inputs weren’t clear.

These waits accumulate across the documents that define product development, requirements, risk controls, test methods, protocols, reports, trace matrices, claims and the justifications that tie it all together. The people who can untangle that web are often the scarcest resources on the program. AI helps when it gets a competent first pass on the table early, so scarce experts spend their time applying judgment instead of drafting and constructing context.

What AI is good at?

1) Getting documents started

AI can produce solid drafts of many work products when it’s given good examples and clear constraints. If it gets you to a 50 percent draft quickly, an expert can often push it to 95 percent in a fraction of the time it would take to start from zero. The first draft is rarely hard intellectually; it’s hard because it’s tedious, detailed and easy to get wrong. Editing a draft is usually faster and less exhausting than building one from scratch.

Teams are usually better at fixing what they can see than identifying what they didn’t think to look for. AI is useful as a structured brainstormer.

2) Fast synthesis when the context is scattered

A common problem is that the “truth” is distributed across email, chat threads, meeting notes, spreadsheets and half-updated documents. Reconstructing context is laborious and expensive.  Feeding this information into AI can enable a 2-4 hour review effort to be completed in 30-40 minutes with comparable or better quality.  These small wins compound when they happen repeatedly across a team.

3) Expanding lists of options

Teams are usually better at fixing what they can see than identifying what they didn’t think to look for. AI is useful as a structured brainstormer.  For example, when exploring how to instrument or characterize a mechanical behavior, AI can quickly list plausible approaches-strain gauges, digital image correlation, instrumented surrogate setups and the tradeoffs worth considering. You still decide what is feasible and defensible. The benefit is a broader option space early, which lowers the chance you discover late that you built the wrong test or wrote the wrong requirement.

4) The virtual reviewer: quality through earlier gap detection

Time with key SMEs is precious and often they are constrained in their capacity.  Using AI to craft virtual reviewers based on their expertise and allowing easy, instant access can help reduce decision cycles and provide useful feedback before the human expert is available to review and provide input.

It’s also a good place to perform basic GDP-style checks by confirming that the report matches the protocol, results align with the stated methods, deviations are documented, and traceability is intact. Used this way, AI becomes a review accelerator that helps real decision makers’ ideas become available sooner.

When AI can be counterproductive

1) Automating the chaos

If the workflow is already sloppy-unclear ownership, fuzzy decision criteria, too many handoffs and vague inputs, AI will increase output without increasing progress. It will give you more drafts, more comments, longer review cycles and more reconciliation work. Individuals may feel more productive, but the system gets more dysfunctional. 

2) Documentation proliferation

It’s tempting to use AI to produce something at every step, such as a plan, a summary, a protocol, a report, a set of meeting notes, a slide deck. But the constraint is often a decision bottleneck, not a writing bottleneck. If AI increases the volume of material that must be reviewed and aligned, it can extend cycle time rather than reducing it.

There’s also a quieter risk. AI produces text that looks clean and confident, which can create false certainty. In regulated contexts, my quick test is simple: ask where a claim comes from. If it can’t point clearly to a standard, regulation, or internal requirement, treat it as suspect.

And beware “AI slop.” The discipline is carving. If AI gives you ten paragraphs, keep three.

Speed with control

AI will not lower the bar for medical devices. It doesn’t remove accountability and it doesn’t replace expert judgment.  It can help teams clear the bar faster by tightening feedback loops, reducing context reconstruction and finding gaps earlier.

Use AI to accelerate decisions and reviews, not to churn out artifacts. That’s the difference between real acceleration and faster churn.