An architect at her laptop in a studio at dusk, asking an AI assistant to build her a carbon tool

Gen AI for SMBs

From vibe-coded idea to live product in days

The Digimasters Vibe Coded PoC to Scaled Product process: fewer apps, better ones, value sooner.

The overview · 3 min 9 s

From a laptop to the whole studio, in days, not months.

Captions are on by default. Narration uses an AI-cloned voice. The people in the video are AI-generated characters and the carbon figures are illustrative.

Read the transcript

Colleague: That’s brilliant. Can I use it on my project? And could we put the client’s data in it?

Architect: I don’t know. But if you think there’s value in this, we can find out.

That’s the moment that matters. Not the app. The question. Gen AI has made building software almost free. Someone in your business has already built something like this. So the question isn’t how many apps your people can make. It’s which ones are worth making properly, and how quickly you can get the value out of them. Most AI pilots never reach production, and plenty that do never pay back. A hundred unsupported apps isn’t an innovation strategy. It’s a maintenance bill.

The return comes from two things. Backing the right idea, and getting it in front of people early. Every week a good idea waits in a queue is value you never get back. So let’s follow her carbon tool through, from a laptop to the whole studio, in days, not months.

She built it in an afternoon, inside the AI policy. That proves the idea is real. Not that it’s ready.

She uploads it. AI writes a specification of what it actually does. She makes the case: every early-stage scheme in the studio could test carbon options in minutes, not days. We score it against your priorities and check nothing like it already exists. Within 24 hours it’s go, rework or park. Most ideas won’t make it through, and that’s the point.

Hers does. Now it’s built properly. A secure repository. A real database for project data. The right architecture for the job, not the biggest one. Sign-on, tests and automated deployment, so it lasts.

And her colleague’s question gets a proper answer. Can client data go in it? We check the data, the privacy and the access, scan the code, put Cloudflare in front and run a penetration test. Quality isn’t a phase at the end. It’s the gate.

The studio tests it, it’s approved, and it goes live for every project team. Then we measure it against the case she made, because value you can’t see isn’t value. What we learn shapes the next idea.

It runs where it suits you, from Vercel and Railway to Azure and AWS, with Cloudflare in front of all of it.

And the Digimasters AI Training Academy builds the judgement behind all of it. How to spot an idea worth backing, how to make the case, and how to own a product rather than just prompt one into existence.

Fewer apps. Better ones. Value sooner. That’s how Gen AI actually pays back. Talk to us at Digimasters.

What's changed

Gen AI has made building software almost free.

Someone in your business has probably already built something useful. What's scarce now is the judgement to know which ideas deserve to be built properly, and the speed to get the good ones paying back.

The problem

Most AI pilots never pay back.

A hundred unsupported apps isn't an innovation strategy. It's a maintenance bill.

Our view

Quality over quantity. Value sooner.

  1. 01

    Back the right idea.

    Every idea gets a fair hearing within a day. Only the ones with a real case go forward.

  2. 02

    Build it properly.

    The right architecture for the job, secured and tested, so it lasts and can hold real data.

  3. 03

    Get it live early.

    Every week a good idea waits is value you never get back. Ship in days, then measure it.

The story

One useful idea.

Her colleague leans over the laptop, pointing at the carbon tool and asking if he can use it

An architect asks an AI assistant for a quick tool to test the embodied carbon of her scheme. By the afternoon, it works.

“That's brilliant. Can I use it on my project? And could we put the client's data in it?”

Her colleague

“I don't know. But if you think there's value in this, we can find out.”

The architect
The scheme carbon check app showing 1,764 tCO₂e, marked over target
Over target1,764 tCO₂e
The same app after swapping concrete for timber and adding a cement swap: 1,455 tCO₂e, 18% lower and under target
Under target1,455 tCO₂e, 18% lower

Illustrative figures.

That proves the idea is real. Not that it's ready for every project, or for client data.

Try the tool she built

Illustrative demo with made-up carbon factors. Drag the materials and watch the total move.

Open the demo (opens in a new tab)

The route

Typically 3 to 7 working days, end to end.

The five stages, Spark, Assess, Build, Secure and Ship and scale, with their target times and gates. The usual IT route takes weeks to months; the Digimasters route takes days.
  1. Spark

    An employee builds a working idea with an AI tool (Copilot, Claude Code, Codex), inside your AI policy

    Who
    Your employee
    Target time
    Hours
  2. Assess

    Uploaded to our portal, specified by AI, given a business case and scored against your priorities

    Who
    Digimasters + the employee
    Target time
    Hours to 1 day
    Gate
    Go, rework or park
  3. Build

    Secure GitHub repo, right-sized architecture, rebuilt to scale with SSO and tests

    Who
    Digimasters
    Target time
    1 to 3 days
  4. Secure

    Scanned, behind Cloudflare, pen tested, and checked for privacy and accessibility

    Who
    Digimasters + Cloudflare
    Target time
    1 to 2 days
    Gate
    Security sign-off
  5. Ship & scale

    Tested by your team, approved, live on secure hosting, and measured against the business case

    Who
    Digimasters + your team
    Target time
    Hours to 1 day
    Gate
    Production approval

Target timings for a departmental tool. Larger or higher-risk apps take longer.

See it work · 70 s

Every step, from upload to live.

The full 14-step process

Open full size (opens in a new tab)
The full 14-step process across the five stages and their three gates, with the hosting lanes it runs on and the AI Training Academy underneath.

Scroll sideways to follow the whole diagram.

Quality

Quality is the gate.

  1. After Assess

    Go, rework or park

    Your decision group weighs the case within 24 hours. Rework goes back to the employee with clear feedback.

  2. After Secure

    Security sign-off

    Fix, retest, then release. Nothing touches client data until it has passed.

  3. Before go-live

    Production approval

    An owner, support and a run budget are agreed before anything goes live.

Most ideas won't make it through, and that's the point.

Right-sized, rebuilt to scale

  • A private, protected GitHub repository from day one
  • An architecture review deciding between a single page and three tiers (front end, API, database)
  • Single sign-on, automated tests, CI/CD and infrastructure as code
  • Simple tools can stay single-page and skip the rebuild

Nothing goes live without sign-off

  • Automated scans of code, dependencies and secrets
  • Cloudflare in front (WAF, Zero Trust, DDoS and DNS)
  • A penetration test against the staging build
  • A data and privacy check covering DPIA, GDPR and accessibility
A padlock glowing in a dark server room

Procurement

How we make an AI-built app ready for procurement.

AI tools build a working prototype fast. Buyers need more than a working prototype. We check your app against more than 50 controls, fix the gaps, and give you the evidence.

  1. Intake

    Nine questions decide which checks apply to your app.

  2. Audit

    Automated scans and a manual review.

  3. Fix

    We fix the red items first.

  4. Evidence

    A red, amber and green report for your procurement and legal teams.

  • Fast at scale

    28 checks

    • Servers in the same region as your data (UK by default)
    • Database queries indexed, paginated and free of N+1 loops
    • Timeouts, rate limits and background jobs, so one slow service or one heavy user cannot take the app down
    • Pages render on the server; images, fonts and scripts load only when needed
    • Core Web Vitals measured on mobile, on a slow network
    • New infrastructure only when measurements show a need
  • Safe to buy

    37 checks

    • A data inventory that drives your privacy notice
    • Trackers blocked until consent, and marketing consent kept separate
    • Export and delete-my-data built in, with retention limits enforced
    • Subprocessors listed, each with a data processing agreement
    • No dark patterns, fake reviews or claims you cannot evidence
    • WCAG 2.2 AA accessibility, and an accessibility statement for the public sector
  • Clear to use

    5 checks

    • Error messages say what failed and what to do next
    • One word for one thing across the app, emails and help
    • Short sentences and active voice, based on ASD-STE100, the writing standard for aircraft maintenance manuals
    • Help articles and handover documents your team can follow
  • Secure

    7 checks

    • No known high or critical weaknesses in the code it depends on
    • Visitors never see internal error detail
    • Security headers on every host
    • Every database query parameterised
    • Sign-in and role checks on every change
    • One customer can never see another's data, and no secrets in the code or the logs

Short films

Three of the pillars, in about a minute each. Captions are on by default.

  • Fast at scale · 69 s

    Will your AI-built app survive 1,000 users?

    The scores are our own page’s Lighthouse results from 2 November 2026. The rest are diagrams.

    Read the transcript

    So you've built a proof of concept, or vibe-coded an app in a weekend. Brilliant! Now let's turn it into a product that scales and adds real business value. Part one: make it fast. Your AI-built app works on your laptop. Will it work for a thousand users on a train? Our checklist looks for faults like these. Servers in the US, with the data in London. A database query inside a loop. An AI call with no timeout, so one slow answer holds up everyone. So we check where the app runs and where its data lives. We index the database and page long lists. Every outside call gets a timeout, and AI features get a limit per person. Then we test it on an ordinary phone, on a slow network. Our own PoC to Product page scored 79 for speed on a phone. After these fixes, it scores 89!

    We add infrastructure only when a measurement shows the need. 23 performance checks. One report. Talk to us at Digimasters.

  • Safe to buy · 68 s

    Your app works. Is it ready to launch?

    The apps and suppliers are invented. The cookie banner is a real capture of our own site.

    Read the transcript

    So you've built a proof of concept, or vibe-coded an app in a weekend. Brilliant! Now let's turn it into a product that scales and adds real business value. Part two: make it safe to buy. Your app works. Is it ready to launch? A buyer's review checks that your privacy notice matches what your app really does. Our checklist catches gaps like these. Trackers that load before anyone says yes. A box asking people to agree to the privacy notice, which is a notice, not a contract. And a compliance badge with no audit behind it. So we start with a data inventory: what you collect, why, and for how long. Your privacy notice is written from it. Then we test the consent banner, build export and delete, list every supplier that handles data, check each has a data processing agreement, and scan for accessibility. We build the controls. Your legal team approves the documents. You get the evidence. Talk to us at Digimasters.

  • Clear to use · 67 s

    Why we write like aircraft manuals

    The screens are invented examples.

    Read the transcript

    So you've built a proof of concept, or vibe-coded an app in a weekend. Brilliant! Now let's turn it into a product that scales and adds real business value. Part three: make it clear to use. Aircraft maintenance manuals use a controlled language, so nobody misreads a repair step. It's called Simplified Technical English, and our writing rules are based on it. Take an error message. "Something went wrong" tells you nothing. "Payment failed: your card has expired. Update it in Billing." That says what failed, why, and what to do next. Five rules do most of the work. Say who does what. Keep sentences short. One idea per sentence. One word for one thing, every time. And no idioms.

    We apply them to buttons, errors, emails, help articles and the handover notes your team will use. Clear words, so people can finish the job without calling you. Talk to us at Digimasters.

Narration uses an AI-cloned voice.

Before and after

  • Invented example

    The booking page

    Bookings were never deleted, though the privacy notice said 12 months. Now a daily job deletes each one 13 months after the visit.

  • Invented example

    The client portal with AI summaries

    A "Store nothing" switch was saved but never read, so every document was kept. Now the switch decides, and a test proves it.

  • Invented example

    The team dashboard

    The list loaded the first 1,000 jobs and quietly dropped the rest. Now the database filters it and returns 50 at a time.

The apps are invented and the figures are illustrative. Each finding is a real check from the list, and the full checklist has all three in detail.

The full checklist

All 77 checks in plain English, with why each one matters and who asks. Rate your own app: Red, Amber, Green or Don't know. As a PDF and a CSV for spreadsheets.

Book a PoC review

We put technical controls in place and prepare documents for your legal team to approve. We do not give legal advice.

Hosting

It runs where it suits you.

  • Vercel

    Rapid prototyping

    Front end
    Next.js or static site
    API layer
    Vercel Functions
    Database
    Postgres via Marketplace (Neon)
    Identity and secrets
    Auth.js, Clerk or Entra ID
  • Railway

    Rapid prototyping

    Front end
    Static site or container
    API layer
    Containers (Node, Python)
    Database
    One-click Postgres or Redis
    Identity and secrets
    Your identity provider and environment variables
  • Microsoft Azure

    Enterprise, M365 clients

    Front end
    Static Web Apps or App Service
    API layer
    Container Apps and API Management
    Database
    Azure SQL or PostgreSQL
    Identity and secrets
    Entra ID and Key Vault
  • AWS

    Enterprise

    Front end
    Amplify, or S3 and CloudFront
    API layer
    API Gateway with Lambda or ECS
    Database
    RDS, Aurora or DynamoDB
    Identity and secrets
    Cognito and Secrets Manager

Cloudflare sits in front of every lane, and every change gets a preview or staging environment before it goes live.

Digimasters AI Training Academy

The judgement behind all of it.

How to spot an idea worth backing, make the case, and own a product rather than just prompt one into existence.

Five modules

  1. Gen AI literacy and safe vibe coding
  2. Writing a business case for an idea
  3. From citizen developer to product owner
  4. Secure-by-design habits
  5. Measuring value and adoption
A trainer leading an AI Training Academy session for a small group with laptops

Roles

Who does what.

Your employee
Builds the first version, makes the business case, tests it and owns the product once it’s live
Your decision group
Decides go, rework or park within 24 hours, and gives production approval
Your IT or security lead
Owns the AI use policy, chooses the hosting lane and signs off security with us
Digimasters
Runs the portal, writes the specification, scores the idea, rebuilds it, secures it, hosts it and measures it

Getting started

Your first month Indicative

  1. Week 1

    Set up

    AI use policy agreed, portal and decision group in place, hosting lane chosen

  2. Week 2

    First ideas in

    Uploaded, specified and scored, first go decisions

  3. Week 3

    Build and secure

    Chosen ideas rebuilt to scale, security and quality checks passed

  4. Week 4

    Live and measured

    Signed off and live, usage and value tracked from day one

Talk to us

Fewer apps. Better ones.Value sooner.

That’s how Gen AI actually pays back.