AI is fundamentally shifting how nonprofits fundraise. For the first time in history, nonprofits have the tools necessary to create truly personalized, N of 1 experiences for their donors – while saving hours of time each week.
Nonprofit teams can now build software, reports, and automation with AI that used to require a team of developers, a project manager, and a large budget. With AI tools like Claude, you can explore your data in plain language, automate routine work, and build custom software in days instead of quarters.
This post is my invitation to start building! Leveraging AI has become an imperative for nonprofits who want to accelerate giving and increase their impact. But building comes with responsibility. Before you start, it’s critical to understand the guardrails needed to keep your donors and your mission resilient and safe.
The following recommendations are designed to help you 1) start building today where it makes sense, 2) reduce risks in your organization, and 3) determine when buying instead of building creates more leverage.
Step 1
Start by exploring your data
The lowest-risk, highest-reward place to begin is a conversation with AI in a Chat interface. Buy the paid version of Claude or ChatGPT. Then, upload an Excel export of your giving history, program outcomes, or survey results, and ask the questions you never had time to dig into:
- Which donors gave three years in a row and then went quiet?
- How do outcomes differ across our program sites?
- What are the themes in last quarter’s participant feedback?
When an answer is useful, ask Claude to turn it into an artifact: an interactive dashboard, a calculator, or a visual summary you can share with your board or team. You don’t need a data analyst or developer to start.
As you work, focus on the key habits that make this type of work safer. Start with exports that are aggregated or stripped of personal details where you can, and agree on an AI use policy before staff import donor records. Anthropic’s Claude for Nonprofits resources include free courses and a tutorial on data handling and drafting an AI use policy. Anthropic also announced discounts of up to 75% on Team and Enterprise plans. In addition, you can find templates for creating an AI Policy at Fundraising.AI.
Step 2
Automate the busywork with tools like Claude Cowork
The next step up is handing work off instead of chatting about it. Cowork and other Agentic AI tools let you describe a goal, point AI at the files and tools you choose, and come back to finished work for your review.
Good first candidates are tasks your team already repeats:
- Pulling data from several sources into a funder report
- Extracting figures from PDFs into a spreadsheet
- Reconciling local exports and flagging what doesn’t match
- Preparing board packets and weekly summaries on a schedule
- Emailing yourself a set of daily KPIs or data changes to stay on top of your work
If you are using Claude, you can choose the folders and tools Claude can work in, and it asks before deleting anything. Start with one recurring task, review the output closely the first few times, and let the hours you save go back to the work that brings you joy.
For more advanced users, experiment with multi-step automation tools like N8N or Zapier. These tools let you visually string together tasks into a workflow in order to replace entire streams of manual work.
Step 3
Build your own apps with Claude Code, Lovable, or Replit
This is where it gets exciting. Tools like Claude Code, Lovable, and Replit let you describe an app in plain language and watch a working version appear out of thin air. Using these tools for the first time is literally like magic. A volunteer shift scheduler, an impact calculator, a program intake form, a grant deadline tracker, or a board dashboard can go from idea to prototype in an afternoon. Even a prototype that never ships is valuable, because it shows your team exactly what you need.
Speed is the easy part. The skill to build is making what you build safe, stable, and maintainable over time. And there are a few principles that should guide your focus.
Build safely: thin UX, minimal data, strong sign-in
This next section is a bit “techy” – so if you don’t plan on building your own apps anytime soon, feel free to scan and skip down to the “Build What’s Unique” section!
Keep the UX thin. Make your app a light front end built on top of systems you already trust, such as your CRM or a Google Sheet, using APIs or MCP connectors. Leave the business logic and the sensitive data where they live today. The less code you own and the less data you store, the less you have to secure and maintain. Building a prototype with AI is easy. But securing, fixing, and maintaining a stable app over time still requires full-time help from engineers who understand the logic. It’s not impossible, but the thinner the app, the better.
Store as little data as possible. Don’t copy donor records into a new database when the app can read them on demand. Collect only what the task needs, and never handle card numbers yourself. Use a hosted payment page and never store financial data. Data you don’t store can’t leak.
Use SSO and MFA. Don’t build your own login. Put the app behind the single sign-on your team already uses, and require multi-factor authentication. Enforce access on the server side, because checks that run in the browser can be inspected and bypassed. For more information, check out Replit’s guidance, which likewise suggests using its built-in authentication where possible.
A few more CRITICAL habits cover most of the common mistakes:
- Keep API keys and other secrets in the platform’s secrets store, never in the code or the browser. Lovable’s security guide treats any secret in frontend code as compromised.
- Turn on row-level security and encryption for every database table that holds personal data, and review the policies early.
- Keep the app private or limited to your workspace unless it truly needs to be public.
- Ask the AI to review its own work against the OWASP Top 10, then have a second model or a person check it. See Engadget’s vibe-coding safety tips for more suggestions.
- Keep dependencies on third-party libraries up to date. AI hackers are now able to exploit unpatched libraries in a matter of hours if your team can’t act quickly.
Build what’s unique to your nonprofit
The best reason for building a new tool is when it solves a problem that no vendor will ever solve for you: the way your program intake actually works, your unique volunteer matching rules, your bespoke funder reporting format, the tool your community needs that exists nowhere else. That is where custom software pays off, because it captures something only you know.
For common needs, such as managing donors, taking payments, and keeping the books, you’re better served by products built by teams who solve those problems for thousands of organizations. Buy the standard features and build the features unique to your nonprofit.
Build or buy: three tests
It makes sense to build when all three of these are true:
- You have someone to maintain it over time. Not just someone who can build it, but a named person (and a backup) who will fix it, update it, and answer questions long after launch. This person should also have a clear understanding of security in the changing AI landscape.
- It’s relatively simple. A focused tool with a clear job is a good candidate. A sprawling system with deep rules and many integrations is not.
- It solves a problem specific to your nonprofit. If thousands of other organizations have the same need, a vendor has probably already solved it.
If any of the three is missing, buy or configure something instead.
Why it rarely makes sense to build enterprise software like CRM with AI
AI has made writing software dramatically cheaper. But it has not made owning software cheaper. Your CRM holds your most valuable data (donor records, giving history, and receipts), and it’s imperative for your data to stay secure, accurate, actionable, and current. That is an operating commitment, not a weekend project.
Here are four reasons we think most organizations are better off not vibe coding their own.
1) Security: attackers have the same (or better) AI tools you do
October 2026 has become known as Black October. This label covers a cluster of unrelated incidents, including critical PHP flaws, a compromised release of the LiteLLM AI gateway, and a compromised certificate authority, plus breaches at nine AI developer platforms where attackers reached stored customer secrets. Several of the PHP flaws were found with AI-assisted research, and Amazon saw an exploitation attempt four days before the patch shipped.
These attacks were also widespread across traditional SaaS companies, and organizations like Virtuous repel hundreds of attacks a week with a top-tier security team and sophisticated AI-powered security infrastructure.
The pace of change here is unprecedented. Former CISA director Chris Krebs noted that roughly 3% of all critical vulnerabilities ever discovered were found in the previous month, according to the same report. Defending against that takes dedicated, cutting-edge security teams.
Since Black October, the attacks haven’t slowed. Every new model makes the attackers better. And it has become increasingly difficult for smaller or unsophisticated teams to catch up.
Vibe-coded apps start from a weaker position:
- Veracode tested more than 100 models and found about 45% of AI-generated samples introduced OWASP Top 10 vulnerabilities, with no improvement across testing cycles.
- Georgia Tech’s Vibe Security Radar traced 74 confirmed CVEs to AI coding tools by March 2026, 35 of them in March alone. Researchers believe the true count is five to ten times higher.
- A June 2026 academic audit found at least one vulnerability in 91% of 200 deployed vibe-coded apps. In February, an AI-built social network shipped its database key in browser code, exposing 1.5 million API tokens and 35,000 email addresses.
Vibe-coded apps without top-tier security oversight are an imminent threat to that data, and your donors are the ones who pay for it.
And security monitoring is only half the battle. When you pay a SaaS company, you are also paying for PCI compliance, SOC 2 compliance, regulatory compliance, monitoring, disaster recovery and backup, and many other related services to ensure that your data is safe, live, and recoverable.
2) Business rules: the hard part was never the code
Vertical-specific CRMs contain thousands of business rules accumulated over years of customer conversations. The code has become easier to write. Understanding how all the data has to fit together for receipting, financial accountability, and industry standards still takes countless hours of human work, and AI can’t shortcut knowledge nobody has written down yet.
Here are a couple of small examples that show how much hides inside a “simple” gift record:
- Receipting. Under IRS rules, a donor needs a contemporaneous written acknowledgment for any single gift of $250 or more. If the donor gets something in return, such as a gala ticket, the charity must give a written disclosure when the payment exceeds $75, stating the value of what was received. The penalty is $10 per contribution, up to $5,000 per event or mailing. The caveat is that when a gift comes from a Donor Advised Fund (DAF), the fund itself needs the financial record of the transaction while the person behind the DAF needs the acknowledgement. Your CRM has to know which gifts trigger which statement and calculate the deductible portion correctly every time.
- Payments. PCI DSS 4.0.1 has been fully enforced since March 31, 2025. It includes universal MFA and, for donation pages, inventory and monitoring of every script that runs on the payment page. Teams that build their own donation flow own that work, along with the annual validation.
And those are only two. A real fundraising CRM also has to handle pledges versus payments, recurring gifts, soft credits, matching gifts, tribute gifts, fund designations, refunds, and chargebacks, and keep all of it coherent when one changes. Each rule looks small. But when combined, they create complexity that requires years of industry expertise and product management experience.
3) Team size: you’re racing people with the same tools
SaaS companies use the same AI coding tools your team has access to, so innovation is accelerating on their side, too. The real question is whether an internal team of one or two can keep pace with a team of 3 to 40 top-tier engineers using those same tools.
The gap is wider than headcount. A vendor team also has dedicated security engineers, QA, support, and product specialists. It spreads the cost of all of them across every customer, and it learns from how thousands of organizations use the product. An internal build gets none of that leverage. It gets whatever the one or two people building it have time for.
Over the next two years, the pace of innovation at Virtuous and other SaaS companies will accelerate dramatically. The organizations that are able to take advantage of tools built by the most sophisticated teams will have an inherent advantage.
4) Maintenance: zero to one is easy; year two is where it hurts
Getting from zero to one with a vibe-coded app is easy. Maintaining it and fixing it over time is really hard. Software is about more than building. It means keeping the system stable continuously, innovating, training your team, supporting users, and providing data services.
The research is starting to back this up. A study of 304,362 verified AI-authored commits across 6,275 GitHub repositories tracked what happens to AI-introduced problems after they land in real projects. It concludes that AI-generated code can bring long-term maintenance costs, and one summary of the study reports that unresolved AI-introduced issues grew from a few hundred in early 2025 to more than 110,000 by February 2026. A separate analysis of code quality trends found AI-assisted commits now make up about a quarter of all commits, alongside rising technical debt on eight maintainability metrics.
A custom CRM also needs a long list of work nobody sees in the demo:
- Patching dependencies and responding to new vulnerabilities
- Backups, disaster recovery, and uptime monitoring
- Keeping integrations working when payment, email, and accounting vendors change their APIs
- Updating receipting and compliance logic when rules change
- Documentation, onboarding for new staff, and user support
- Data migrations whenever the model needs to change
Then there is key-person risk. When the one person who understands the system leaves, so does the knowledge of how it works. Which brings us to a story close to home.
A real, personal example
I started my career building a CRM from the ground up for a nonprofit, with a team of three. And it worked relatively well…until the team left. Once the people who understood the system were gone, the nonprofit could no longer maintain it or keep innovating, and it had to switch to an off-the-shelf solution.
That was before AI coding tools. Code is far cheaper to write today, but people still leave, and the knowledge of why the system works the way it does leaves with them. And the market still shifts exponentially faster than most organizations can hire, pivot, and innovate.
Where building does make sense
Hopefully, my passion for building using AI was clear at the start of this post. The time to build is now! At Virtuous, we are building a future that encourages organizations to build on top of our platform. We are releasing an alpha version of our MCP this year and have a goal of making the entire platform agent accessible.
AI tools are great for lightweight tools built on top of stable, AI-powered platforms. They are also great for one-off reports and dashboards, automations and integrations that sit around your CRM, and prototypes that help you work out what you need. The line is between building on top of a trusted system of record and replacing it. Vibe code on your CRM, not instead of it.
Before you build, ask these six questions
- Who is accountable for security, and what happens when a vulnerability is disclosed on a Friday night?
- Who knows every rule behind a receipt, a refund, a recurring gift, and a PCI audit, and where is that written down?
- What happens to the system when the person who built it leaves?
- Who will train new staff, answer support questions, and fix the bug found at year-end?
- Can a team of one or two keep up with a vendor team of dozens using the same AI tools?
- What is the full cost of ownership over five years, including the time you’ll spend maintaining it instead of pursuing your mission?
If you can’t answer all six with confidence, a vibe-coded CRM is likely an untenable risk to your donors and your mission. But that shouldn’t stop you from building. Start small, start lightweight, focus on security first, and scale up the features that make the most unique impact on your mission!


