)
)
Key takeaways
Not developers, but builders anyway: Our Support team don’t write code, but they've built three working AI tools from scratch by talking to Claude in plain English.
Their favourite tool saved three hours a day: An alert-triage tool that used to take up to 30 minutes per user now gives an answer in one click, cutting a typical post-weekend triage from three hours to fifteen minutes.
They build the boring bits away: The goal was never to replace the team, it was to hand the repetitive slog to the tools so the humans get more time for the work that actually needs them.
It takes time to trust the tools: The tools got good by running alongside the manual process, correcting Claude, and teaching it the "why" until it stopped going rogue.
The real blockers aren't always technical: The hard part wasn't the building. It was access, permissions and keeping an eye on costs, the stuff that's trickier when you're not a developer.
Meet Jade and Winter from our Support team. When Cube went AI native, these two jumped in and went HAM building. Here's what they made and how they made it without writing a single line of code.
Meet our Support team
Every team at Cube has felt the Factory Rebuild differently. The reactions range from existential menty b’s, to taking to this new world order like a duck to water. Our Support team went full duck mode.
Jade and Winter are our ridiculously talented Cubes who make up our Support team. But what do they actually do in Support?
To cut to the chase, they're the ones who keep things running after a project goes live. They triage tickets, chase bugs, monitor how projects perform in the wild, sort out issues before multiple users even notice them, and they report back to clients on all of it.
A lot of it, historically, was manual and repetitive. But hold that thought!
Neither Jade nor Winter are developers, and are self-proclaimed ‘non-technical goddesses’. Neither of them writes code. Which makes what they've built over the past few months pretty damn impressive 👀
)
Why they didn't hit the panic button
When Duncan announced 3 Sided Cube was going all-in on AI, it raised plenty of questions and a couple of wobbles. It's been a running theme across this whole series that the shift took some time to adjust to. But Jade and Winter? They were weirdly calm.
For Jade, it was a case of been-there-done-that. Her background in startups meant she sees sudden pivots as a regular Tuesday. Plus, she'd already worked somewhere that built "an army of robots" to handle the repetitive stuff, so she wasn’t worried about being replaced any time soon. Instead, the idea of handing the boring jobs to automation was familiar and super exciting.
"One of the most exciting things for me was knowing we weren't going to lose our jobs, we'd shift from being on the tools to more of a prompt-engineer role. Get the robots to do the on-the-tools stuff, so we could do the bespoke, high-value delivery."Jade Reynolds, Support Admin, 3 Sided Cube
Then, Winter came at it from the opposite direction with no startup background and, by her own admission, being less used to things changing at the speed AI has been moving. What settled her was something Duncan said: you'll always be scared of a thing until you actually do it.
"I kind of just wanted to run with it, and use it in the best way to free up my time to learn other things and actually develop more, rather than being stuck in the same jobs every day." Winter Tilley, Junior Support Manager, 3 Sided Cube
By handing the grunt work to AI, Jade and Winter freed themselves up for the stuff that really needs a human touch. They were trusted to experiment and see what stuck, which, as it turns out, is exactly the condition under which people build the most useful things.
The tools they built
The thing about the day-to-day manual work is that you get so used to it and stop noticing how mad it is that you spend so much time on one thing. Jade and Winter's job was full of these things. Multi-step processes all done manually, all important. Another layer of importance was also the real time pressure as a lot of our Support work is on life-saving projects where people are relying on the app in an emergency.
So when they got access to Claude, they didn't build tools for the novelty of it. They built them because doing every task and every step manually was costing their users time they didn't have.
)
The Alert Triage tool
Just to set the scene, someone using a disaster-preparedness app gets in touch, worried. "My friend in the same town as me got an alert about a weather event and I didn't. Should I have? Is your app broken? Am I in danger?" As Winter explains, not getting an alert can be just as alarming as getting one.
Answering what seems like a straightforward question used to be a long, long LONG process:
Take the user's monitored locations (which could be anywhere between one and twenty),
Check three different websites for the weather event happening in those locations,
Plot each set of monitored location coordinates on a map,
Pull up the alert zone on another map,
Line up the two maps by eye, zooming in and out to see whether the user fell inside or outside the warning area,
If they shouldn’t have received an alert, respond to the user and reassure them,
If they should have received an alert, respond to the user and make sure they’re safe before investigating,
It may be a settings-related issue on the users’ end, like having notifications turned off for that event type, or maybe they had critical alerts disabled,
Or it may be caused by a bug, in which case a ticket needs to be raised and resolved.
Per user, the process took ten to thirty minutes. And in a big weather event like tornadoes or an earthquake, hundreds of users might be asking all at once.
Now, the Alert Triage tool does all of that at the click of a button.
All Jade or Winter have to do is drop in the user's data, and it instantly tells you if they were eligible for that alert or not. If they should have got it but didn't, the tool works out why. And if their settings were all correct and they still didn't get the alert, the tool flags it as a genuine bug and hands a developer a ready-to-investigate ticket.
So how did a non-developer build it? Winter more or less talked it into existence.
"I didn't know how to make anything live. I basically went into Claude and brain-dumped a massive ‘this is everything we do, this is the problem, here's all the logic I know’ message. I typed into Claude for about half an hour, and it spat out an HTML file you had to save locally on your device."Winter Tilley, Junior Support Manager, 3 Sided Cube
Although that first version only ran locally on Winter's laptop, it's since become a live site any Cube can log into. The tool holds all the context, so you don't need to be an expert to use it, which is a god-send when it comes to holiday cover and generally getting more Cubes involved when needed.
"Winter and I were able to be on holiday at the same time, and Project Managers picked up the process with no friction at all. Before, I'd have looked at the schedule, seen us both off, and gone ‘Crap, that's going to be a painful day for everyone else’."Jade Reynolds, Support Admin, 3 Sided Cube
To put the impact of this tool into perspective, a typical post-weekend triage that used to eat about three hours of Jade's morning now takes fifteen to twenty minutes!
The Extreme Weather Dashboard
When our COO Sophie pinged Jade a Slack message asking why there'd been a spike in tickets on a disaster-preparedness app, Jade honestly didn’t know in the moment and hated not having an answer.
That’s because a spike in tickets could mean a hundred things, and working out which meant cross-referencing Zendesk numbers with Google Analytics as well as checking what the weather happened to be doing at the time. Also, remember when we said you have to check three different websites for weather events as part of the manual Alert Triage process? That’s because there’s no central place for all global weather events either.
Needing all of that information but having no single place that pulled it all together is the opposite of helpful when you’re trying to get to the root cause of an issue.
Jade’s stomach-drop moment of not being able to give Sophie a quick answer was what sparked the build of this tool.
The Extreme Weather Dashboard lines up tickets, active users and extreme weather events in one filterable view, so you can see every piece of important context at a glance. It even filters out the distracting noise (the "it's raining outside" and "my cat needs the vet" tickets that aren't really issues).
Now, the team can see when there’s ten times the usual issues, because there’s ten times the usual users, because there's a hurricane.
Another huge benefit is how proactive the dashboard has made the Support team. It reads the weather forecast and gives the 24/7 Support weekend team a heads-up when a big event's coming, so they go into the weekend knowing what to expect.
)
Where it gets tricky
Building your own software when you can't read code is, as you'd imagine, not entirely smooth. With the tools they’ve created so far, Jade and Winter have some top tips for other non-technical goddesses.
1. Learning to trust what you can't read
The first thing you have to get your head around is trust. When Claude spits out code, a non-developer can't read it to check it's right, you can see that it works, but you can’t see why. That means your trust in Claude is earnt over time rather than with a speedy code review.
"There was a lot of hand-holding, a lot of comparing and giving Claude the context of 'this is why what you did was wrong.' It's an ongoing learning journey for Claude as well as us."Jade Reynolds, Support Admin, 3 Sided Cube
Jade ran the manual triage process alongside the Alert Triage tool for weeks to make sure it was as accurate as possible. All that correcting and explaining her reasoning behind triage decisions paid off. Eventually, it stopped needing any correcting at all.
2. Sometimes the blockers are human
The less glamorous blockers aren't about the AI at all. As a non-technical person, you're not automatically set up with the permissions, accounts and organisation access a developer gets as standard. This meant that access was a bit of a hurdle.
Jade and Winter are the first to say those guardrails are a good thing, you don't want just anyone able to deploy to live projects! But it does mean that when you're on a roll building something useful, you can hit a wall, and then you're back to asking someone for access. Usually our poor Engineering Manager Kev.
"Bless Kev, he's our guardian angel. We just ping him all day, every day, ‘Can you give me this’, ‘Can you grant access to that’."Winter Tilley, Junior Support Manager, 3 Sided Cube
3. The cost you can't see
The same goes for the cost of AI. Without clear visibility, it's hard to know whether you're spending sensibly or, as Jade puts it, quietly funding the demise of the Christmas party open bar.
“Is it dollars? Is it cents? I'm testing this thing, and each time it's costing the business money, I just don't know how much."Jade Reynolds, Support Admin, 3 Sided Cube
Both of them would happily take more oversight there, not fewer guardrails, just a clearer view of the AI Spend O’meter. For now, though, that visibility just isn't there. Rather than trying to solve it themselves, they're happy to leave the spend-tracking side to other Cubes and keep their energy on the tools that are actually theirs to build.
)
Would they go back?
We always ask this question for our Factory Rebuild series. Jade and Winter were super quick and clear on their answers.
"Absolutely not. You couldn't pay me enough money to go back to manually doing all of this. Now that I've seen what's possible, I really pity my March self."Jade Reynolds, Support Admin, 3 Sided Cube
“No way. Now it makes me question everything else we do as standard, I want to tackle this, and this, and this."Winter Tilley, Junior Support Manager, 3 Sided Cube
Neither of them see the tools we’ve talked about as the finish line. They’ve got even more ideas and no doubt those will continue to build up as more and more becomes possible with AI.
Jade's already building the next tool, working through every manual job she still does with the same question of ‘How do I get Claude to do this for me?’. Now, Jade and Winter have stopped seeing the manual grind as just "the job", instead it's a problem to be solved. And now they've got the AI tools to solve it!
Straight from the Cubes
Want more of the low-down on how the Factory Rebuild has gone across the teams ? Get the stories straight from the Cubes and catch up here:
You shall not pass, AI: meet the Gandalf of our Factory Rebuild
How a marketing team goes AI native (in an office full of developers)
Questions clients really ask when their software agency goes AI native
Disclaimer: We used AI in the process of writing this blog, but a human was still at the centre! We'd be poor advocates for going AI native if we didn't use AI to help write about it. The story is ours, the quotes are our team's own words, and every line got the human pass. 💚
FAQs
Can non-technical people really build software with AI?
Yes, but there are limits! You don't need to read or write code to get something live, but you do need to understand the process you're trying to automate and what ‘good’ looks like for your specific use case. If you want the software to be useable by multiple people, secure, and accessible, you need a developer in the loop.
How much time can AI automation actually save?
It varies hugely by task, but as one example, a triage process that used to take our team up to three hours now takes fifteen to twenty minutes. The biggest gains come from automating repetitive, multi-step manual work.
What are the hardest parts of building with AI as a non-developer?
Aside from the actual technical side, the hardest parts were trusting output you can't read, getting the right access and permissions (which exist for good security reasons), and keeping visibility on costs.
Should AI-generated work still be checked by a human?
Yes, especially anything client-facing or safety-related. Our team keeps a human review step in their reporting process because AI can word things in misleading ways or make small errors.
Published on 17 September 2026, last updated on 17 September 2026