The Era of Personal Software. Using AI to build tools that do exactly what you need.

For most of the history of computing, if you wanted software that did something specific to your work, you either bought a product that did roughly what you needed or you hired a developer to build exactly what you needed. The first option meant compromise. The second meant cost, time, and the risk that what you got back was not quite what you asked for. For most small and medium-sized businesses, neither option was attractive, so you made do with whatever was available and filled the gaps with spreadsheets.

That constraint is dissolving. We are entering an era where a professional who understands their business problem clearly can commission software that does precisely what they need, built in hours rather than months, at negligible cost, not by learning to code, but by describing the problem well enough that an AI coding tool can build it for you.

Two examples from our own work illustrate what this looks like in practice.

ChangeSchool operates across Armenia, Georgia, Kazakhstan, and the UK. Our team flies regularly between Tbilisi, Yerevan, Almaty, and London. If you have ever tried to search for flights across Central Asia and the Caucasus on a commercial booking site, you will know the results are poor. Regional carriers are underrepresented, nonstop routes are missing, and the prices shown are often not the best available.

So we built a flight search tool. It queries three separate data sources in parallel: Duffel, which covers over 300 airlines with direct connections; Amadeus, which reaches hundreds of regional carriers through its global distribution system; and Google Flights via a search API for gap coverage. The results are merged, deduplicated, and enriched with independent airline safety ratings from IATA, the European Aviation Safety Agency, and AirlineRatings. The whole thing runs from the command line, exports the results in an Excel file, and a search costs effectively nothing.

The results speak for themselves. In a recent search for Tbilisi to Almaty, our tool found an Air Astana nonstop flight at GBP 184 that did not appear on Skyscanner or Kayak. It also surfaced connecting options via Etihad and Lufthansa that only existed in the Amadeus global distribution system and were invisible to consumer booking sites. For a company whose team regularly crosses these routes, that is a practical, measurable difference.

The second example is less glamorous but arguably more useful. Every company that uses a CRM accumulates data quality problems over time. Duplicate contacts appear when people are entered from different sources. Contacts lose their connection to company records. Regional assignments drift as the business changes. In a large company, a database administrator handles this. In an SME, nobody does, and the data quietly degrades until someone notices that the pipeline reports do not match reality.

We built a HubSpot assistant that handles the CRM administration work nobody wants to do. It detects duplicate contacts by matching across names, email addresses, and phone numbers, then merges them safely with full backups and verification at every step. It finds orphaned contacts and links them to company records using email domain matching. It detects countries from phone prefixes and email domains, and automatically assigns regional owners. It generates data quality reports showing fill rates and missing fields across the entire database.

In its first serious use, it merged 584 duplicate contacts, reducing duplicate groups by 92 per cent, and linked 48 orphaned contacts to the correct company records, creating 26 missing company records in the process. Work that would have taken someone days to do by hand was done in minutes, with safety checks that a human would not have had the patience to apply consistently.

Both of these tools were built in hours using publicly available APIs and standard libraries. Neither requires special hardware or a subscription. Neither was built by a professional software developer in the traditional sense. They were built by someone who understood the business problem, clearly described it, and used an AI coding tool to produce the result.

For clarity, I am not saying we don’t need HubSpot. We absolutely do. It’s a professional-grade, well-maintained database, and earlier in this series, I wrote about how we use that data in our own dashboards and sales support. You also need someone with a little bit of understanding of how APIs work, particularly security and scopes, and how databases work and relate records at the backend. This is to make sure that what you’re asking the AI to do is something you should be asking it to do. You can educate yourself on this, either alone or with the support of the AI, in less than a day.

Also, there is a risk worth being honest about. The Amadeus Self-Service API, which our flight tool depends on for regional carrier coverage, is being taken offline on 17 July 2026. We knew this was possible when we built it, and we have already integrated alternative data sources to cover the gap. But it is a real illustration of the dependency risk that comes with building on third-party APIs. When you create personal software, you accept that some of the services it connects to will change or disappear, and you need to be prepared to adapt. That is a management consideration, not a technical one [1].

The broader pattern extends further than these two examples. We have also processed 50,000 photographs on a standard office machine using local AI models for image indexing and website accessibility text, with no cloud dependency. None of these projects required a development team or a significant budget. What they required was clarity about the problem and enough understanding of the output to review it critically.

The five-step discipline from earlier in this series applies here as directly as anywhere. Define the business need: we cannot find reliable flight options across our operating region. Define the outcome: a search tool that queries multiple sources and returns comprehensive results with safety data. Review the plan, delegate the build, and review what comes back. The human decides what to build and whether the result is good enough. The AI does the building.

If you have followed this series from the beginning, the thread is consistent. Your data, your interface. Your inbox, your rules. Your documents, your knowledge base. Your problems, your software. Any professional can now think in these terms. Not applications you buy off the shelf, but tools you commission for your exact situation. The technology is available to anyone with a clear idea of what they need.

---

Sources:

[1] Amadeus for Developers Self-Service API: end-of-life 17 July 2026; new registrations already paused

This article is part of a series on practical AI applications at ChangeSchool. For the strategic view of leading AI in organisations, see Viren Lall’s companion series. ChangeSchool is an SME without a dedicated coding resource. Every application described in this series was built using AI coding tools. They are all vibe coding projects, and all within the reach of a startup or a micro business willing to spend some hours and have a go. If you’ve got any questions about what this was or how we did it, or any suggestions for better ways forward, please pop them in the comments below, and I’ll do my best to answer them.

Technical: Our process is the same for every project. Once the business need and objective are defined, we research the approach using Google Gemini Deep Research and Perplexity Deep Research, then give that research to Claude Code to build it. None of the technical terms below were things we knew beforehand; they came out of that research process. It takes minutes to get up to speed. All built in Python using Claude Code. The flight search tool uses the Duffel API (duffel.com), Amadeus Self-Service API (developers.amadeus.com), and Google Flights data via fast-flights and SerpApi. The CRM assistant uses the HubSpot API (developers.hubspot.com). The photo indexing mentioned in this article uses InsightFace for face recognition, ARNIQA and LIQE for image quality scoring, and OneAlign for aesthetic assessment (huggingface.co/q-future/one-align). All image models run locally on consumer hardware and can be downloaded free from Hugging Face (huggingface.co).