On 25 August 2026, OpenAI announced the eight winners of Build Week, a hackathon built around Codex and GPT-5.6. Nearly 47,000 builders from 186 countries took part, submitting more than 8,000 projects over eight days. Two winners stand out for anyone running an Australian business: a veterinarian with no coding background who built a triage tool now being piloted in her short-staffed clinic, and a cardiologist in Cairo who built a research prototype for cardiac-arrest response between hospital shifts.
Set aside the hackathon framing for a moment. What those two people did is the thing most businesses have been trying and failing to do for a decade: get the person who understands the problem to build the thing that fixes it, without a six-month project and a development budget.
Can someone with no coding background actually ship a working tool?
Yes, and the Build Week winners are the clearest public evidence to date. A practising veterinarian with no coding background produced a clinic triage tool that moved from idea to a real pilot in her own workplace. A working clinician produced a research prototype around cardiac-arrest response in the gaps between hospital shifts. Neither had a development team, a sprint board or a budget line. Both had deep knowledge of a specific problem and an AI tool that could turn that knowledge into something that runs.
That is a meaningfully different claim from the usual no-code pitch. The tools were not generic dashboards assembled from templates. They were built around a domain expert's understanding of what the work actually requires, which is the part that normally gets lost in a handover to a development team.
Why the domain expert is the bottleneck, not the developer
In most internal software projects, the expensive part is not writing the code. It is the translation loss between the person who knows the work and the person who can build. Requirements get written down, misread, negotiated, descoped, and what ships is a version of the problem rather than the problem. We see this constantly on client work: the operations manager who could describe the correct process in four sentences ends up with a tool that does not match it.
Remove the translation step and the economics change. A Brisbane practice manager who knows precisely which patient flags matter can build the triage logic herself, test it in a week, and throw it away if it is wrong. That last part matters more than it sounds. Cheap discard is what lets a business try five ideas instead of committing to one.
The expert holds the edge cases in their head already, so they never need writing down and handing over.
Feedback loops shrink from weeks to hours, because the builder is also the user.
Failed attempts cost hours, not a project budget, so more things get tried.
The tool stays close to the actual process, and changes when the process changes.
What gets built is small and specific, which is usually the kind of thing that survives.
Where Claude Cowork fits for a non-technical builder
Codex and Claude Code both live in a developer's world. Cowork is the surface built for someone who is not a developer and does not intend to become one: you describe the work, connect the tools the work already lives in, and the agent does it. We have covered the broader capability comparison in our piece on Claude Cowork versus OpenAI Codex and the push beyond coding, so this is the narrower question of who is holding the keyboard.
For a domain expert, the practical difference is what you have to know before you start. A hackathon winner with no coding background still had to work inside a builder's tooling. The Cowork framing assumes no such step, which is why most of the people we onboard are office managers, bookkeepers and practice managers rather than engineers. Our guide to onboarding a non-technical team onto Cowork walks through the first fortnight, and writing your first Cowork skill without a developer covers the step after that.
Two paths to the same outcome, and what each one asks of the person building
| Dimension | Codex style coding agent | Claude Cowork |
|---|---|---|
| Who it assumes you are | Someone willing to work in a developer environment | Someone who knows the work, not the tooling |
| What you produce | An application you then have to host and maintain | A procedure the agent runs against your existing tools |
| Where it runs | Wherever you deploy it | Against the systems the work already lives in |
| Who maintains it | Whoever can read the code later | The person who wrote the instructions |
| Best fit | A standalone product idea | Work that already happens every week |
What this does not mean
A hackathon is not a production system. A triage tool being piloted in one clinic is an encouraging signal, not a licence to put an untested agent between a patient and a clinical decision. In regulated Australian settings, and health is the sharpest example, the build is the easy part. Clinical governance, Privacy Act obligations, record keeping and the question of who is accountable for an output all still apply, and none of them get faster because the tool took a weekend.
Our rule on client work is simple: a non-technical builder owns the first version, and a review gate sits between that version and anyone outside the team. We typically budget $12,000 to $25,000 for the governance and rollout work around a set of internally built tools, which is a fraction of what a bespoke build costs and is the part that actually decides whether the thing survives contact with the business. The ROI calculator will give you a rough shape for your own numbers.
The takeaway for your team
The interesting result from August 2026 is not that a hackathon happened. It is that the winners included people whose expertise is animals and hearts rather than software, and that what they built was real enough to pilot. If you employ people who know your work that well, the constraint on internal tooling was never their ability. It was the tooling in between.
If you want to work out which of your processes a domain expert could own directly, that is the conversation we have most weeks. See our consulting services or read OpenAI's own writeup of the Build Week winners for the full list of projects.
