AI coding tools are helping some nonprofits build or customise software without a traditional development budget. The opportunity is real, but so is the need for human oversight when the software supports essential work.
Watch Desk analysis
What happened
In an article published on 14 October, Allison Fine describes nonprofits using generative AI and agentic coding to tailor software to their needs. Fine is president and CEO of Every.org, and her examples include her own organisation as well as other nonprofits.
The Adventure Project’s co-founder, Becky Straw, used Anthropic’s Claude to write code for a custom map on the organisation’s website. Fine also describes a custom AI agent built for the Blue Devils Drum & Bugle Corps: it organises daily bingo transaction spreadsheets for QuickBooks and flags anomalies. Fine says the system took fewer than 20 hours to build and has reduced outsourced accounting costs by as much as 25% per month.
GiveDirectly built an AI-powered scenario-planning tool to help staff weigh factors such as need, funding and operational capacity. Fine says it informs the decision, rather than making it. She also says agentic coding has helped Every.org increase donations tenfold without increasing its engineering headcount. That last figure is her account of her own organisation’s results, not an independently established measure of AI’s effect.
Fine recommends three starting points: set human-centred standards and policies; identify which systems are genuinely mission-critical; and work iteratively, deciding whether staff can customise a tool themselves or need outside help. She says people should test and review AI-generated code before it is put into use. Her article in Stanford Social Innovation Review also cautions that this is not a reason to abandon commercial software altogether.
Why it matters
For organisations facing high software costs or limited development capacity, AI coding tools may make useful customisation more attainable. The examples range from a website feature to bookkeeping support and crisis planning, not a magic replacement for every vendor or developer.
The more consequential the system, the less sensible it is to treat generated code as ready to trust. Fine’s advice to start with a defined need, keep people responsible for decisions and check the work is especially pertinent when software touches finances or services for people in crisis.
Our read
This is a grounded case for selective independence, not a declaration that every nonprofit should build its own software. The most convincing examples connect a specific operational problem to a specific tool. Start with one manageable job, keep a person accountable for the result, and ask whether the time and support needed to maintain it still make the project worthwhile. “We can build it” is a useful new option; it is not the same as “we should”.
What to watch
- Whether the reported savings and donation growth are sustained and measured against clear baselines.
- How nonprofits test, maintain and secure AI-generated code after the first version works.
- Which mission-critical systems organisations choose to customise, and which they decide to leave to specialist providers.
Discussion spark: If a nonprofit can use AI to build a tool tailored to its work, when is that independence worth the extra responsibility of maintaining and checking it?
Sources and evidence
- Source update (14 October 2026, 12:00 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.