Where AI creates new offerings, not just savings
If your AI roadmap is a list of tasks to make cheaper, you're using half the tool. The more useful question for a leadership team right now is where AI lets you offer something you couldn't offer before, because that's where the durable advantage sits. Efficiency gains reset everyone's baseline and quickly become table stakes. The teams that also hunt for new capabilities are the ones likely to pull ahead.
This distinction was framed recently as efficiency AI versus opportunity AI. Efficiency AI helps you do your existing work faster, cheaper, or with fewer errors. Opportunity AI unlocks work you never previously considered. The framing gained fresh relevance because of how the latest generation of models behaves. GPT-6 Astra, by several practitioner accounts, is an odd release: unusually strong at things like video editing, 3D modeling, and interactive builds, while some coding users reported it felt like a regression and switched back to older models. In other words, the clearest value of a new model is increasingly not in doing your current tasks better. It's in the tasks that weren't on your list at all.
Efficiency and opportunity are different budget decisions
These are not rival philosophies, and efficiency AI will carry most of the near-term value. The point is that they lead to different investments. Efficiency work optimizes a process you already run. Opportunity work asks what your product or service could become if a capability that used to require a specialist team were suddenly cheap enough to try on a weekend.
The strategic risk for executives is treating AI purely as an efficiency technology. Every competitor is already chasing the same cost savings, so those wins commodify fast. The offerings that redefine a category tend to come from somewhere orthogonal to the existing business, which is exactly the space that a cost-cutting mindset never looks at.
What opportunity AI looks like in practice
The thought starters that came with this framing are concrete enough to test. A few worth sitting with:
Interactive proposals clients can shape. A proposal already contains a hidden model of scope, resources, and timeline. When a client asks whether you can finish sooner or add a department, you rerun part of that model by hand. Making selected parts of that reasoning something the client can manipulate lets them explore tradeoffs directly. This one is interesting because it's efficiency and opportunity at once: a genuinely new way to interact with clients that also cuts the back-and-forth latency out of every negotiation.
Custom video production pipelines. Rather than a dedicated clipping product, a coding agent can build a pipeline that ingests a script and raw footage and produces finished clips against a visual style you define. The homework version is small: write a 60-second script, record it on a phone, hand it to a coding tool, and ask it to design a reusable production pipeline. The question that follows is where video would help in your work if the barrier to making it dropped to almost nothing.
3D and simulation. A standout capability of the newer models is operating in three dimensions, from walkthroughs to learning objects you can rotate and inspect. Alongside that, simulated practice environments let people rehearse difficult conversations or decisions and get feedback mid-stream, which is one of the harder things to build into training.
Expertise as a product. Much of expert help is a conversation: gathering context, spotting patterns, ruling out attractive but wrong options. Turning part of that judgment into something people can work through themselves scales your expertise beyond the hours you can personally give, and inside a company it often becomes efficiency AI too, since so much expert time goes to repeating the same guidance.
Getting past the blank page
The hard part is that opportunity AI asks you to consider capabilities you haven't considered, which is close to a contradiction. People don't carry a full inventory of what they could build. We carry a smaller one shaped by our job, our tools, and what we see colleagues doing.
The most useful shortcut offered here is to look sideways. Notice what people in other roles are doing with AI that strikes you as genuinely cool, then ask what the equivalent would be in your work. Many of these opportunities are things other people could already do and you couldn't, until the cost of trying collapsed. And not everything will land. Building a compelling game or interactive experience is still a craft, and plenty of professionally made ones fail. The shift worth naming is that trying is now cheap enough to be worth doing.
My take: Run both tracks, but budget them separately, because if opportunity work competes with efficiency work for the same hours it always loses to the safer number. Pick one client-facing artifact this quarter, a proposal, a demo, or a training module, and rebuild it as something interactive rather than static. Treat the failures as the cost of finding the one that changes what you sell, and ignore anyone telling you a single model is uniformly better or worse when its value now depends entirely on which task you point it at.