AI has taken over my life.
Over the last year, I have facilitated workshops to train hundreds of investors on how to build AI agents, worked with fund managers and entrepreneurs on AI transformation, lost sleep and ignored my friends playing with Claude, and led the charge at the Sorenson Impact Foundation to integrate AI across most layers of our operations.
It’s been a journey driven by the urgency of the moment and the need to increase our capacity to meet it. And one that acknowledges a fundamental truth: AI is poised to transform how our society operates. To meaningfully engage as investors amidst this AI transformation, we need to understand and use the technology ourselves.
The foundation issued a Responsible Technology RFP earlier this year soliciting proposals from startups, nonprofits, researchers and ecosystem builders innovating in AI safety, strengthening the social safety net, and promoting worker resilience. Internally, we have also integrated AI across our core operational processes.
We have built autonomous agents that help with scoring grant applications that uniformly adhere to our scorecards & sped up review. Agents gather and enrich our CRM systems with press mentions and board materials from our investees to support portfolio monitoring. With the help of AI, we collate a monthly list of open positions across the portfolio that we blast to our network to support talent acquisition. Our agents are also constantly scouring the web, sourcing opportunities and sharing deal-flow with our co-investors.
This has helped us deepen our relationships in the ecosystem, accelerate onboarding for new team members, and free up time to engage more deeply on thesis building and supporting investees.
We use just three tools to accomplish all of this.
Our AI note-taker helps us surface trends in meetings and draft follow-up emails. We use no-code workflow builders to create automated workflows that integrate AI. And we use large language models, such as Anthropic’s Claude, to interact with and extract insights and create charts from data across our entire file system.
I’ve compiled six key insights for other investors embarking on their own AI transformation.
Cultivate a culture of continuous learning and experimentation
If you’re a leader, encourage your employees to use AI. Pay for their tools, allow them to experiment and make mistakes, ask them to teach you, provide them with a platform, and most importantly, give them the psychological assurance that adopting AI will not result in them being made redundant.
Accumulate small wins
The single biggest mistake I see investors make when beginning their AI journeys is trying to overhaul their entire firm in one fell swoop. Everyone wants to build a chat bot that is fully integrated across their entire file system. This is a great project that is destined to fail if it’s your first. My simple heuristic for AI change management is that the more stakeholders you have to convince to deploy AI (IT, legal, firm leadership), the more likely you are to fail.
Instead, start small and build agents that give you individual leverage. The first agent I ever built would prioritize my weekly to-do list. The second agent generated a custom newsletter of trends I was following. The third agent scraped the latest social media updates from the foundation’s portfolio companies and wrote an overview once a month.
These agents are easy to create, helped me build my own confidence and gave me quick wins to push for broader adoption. Most importantly, they exist outside of our internal file storage system, sidestepping compliance concerns (more on that next).
Don’t be intimidated by compliance
AI compliance and safety can generally be broken down into three categories: Preventing data leakage, safeguarding cyber security, and reducing human error.
A widespread concern with LLMs is that, if you share your proprietary data with them, it will be leaked externally. To mitigate these concerns, review the Terms of Service section on model-training for your preferred tool and affirmatively opt-out of model-training. Consider purchasing an enterprise plan with enterprise-grade assurances, and/or send an email to the vendor requesting they confirm in writing that they are not using your internal data for model training.
On cybersecurity, every organization has a checklist and set of regulatory requirements on record-keeping and storage that they use for vendor due diligence, and this same framework should be applied to any AI vendor as well. At a minimum, requiring a Soc2 certification, a generally accepted cybersecurity report card for software companies, is a good first step.
Mitigate human error
My biggest concern with AI use, and one that gets too little air-time, is human error. AI agents can act autonomously and process requests at scale, making small mistakes balloon. I’ve had an AI agent send 30 emails to one person (they haven’t responded to an email from me since).
To mitigate human errors, I recommend a proactive “pre-mortem” whenever you adopt a new tool to help avoid adverse outcomes. What are your top five concerns about using the software solution?
For example, are you scared of waiving your attorney-client privilege by having an AI notetaker on during a call with your lawyers? Then turn your notetaker off or have it ask permission to join calls — and use that nudge to say no. Worried that AI will accidentally delete all the files in your SharePoint? Then block its ability to delete files and introduce a human-in-the-loop step when it manipulates records. Concerned with AI making a biased funding recommendation? Ask AI to explain its reasoning and audit its analysis for consistent patterns and biases. And so on.
Don’t outsource AI transformation
It may be tempting, but don’t have an intern or a consultant drive your AI transformation. Investment funds need staff who are spending at least 20% of their time on AI upskilling.
The consensus is that AI is a technology that benefits mid-level employees the most because they have the “context” of an organization. I would extend this argument and say that “systems-thinking” — the ability to see across an organization, deconstruct it into interrelated processes, identify silos, and see how people interact with technology and one another — is essential to building agents.
Only someone who has lived and breathed these processes can find ways to automate and maintain them.
Get comfortable with the discomfort
The technical barrier to building agents has fallen exponentially. In fact, the biggest barrier that people face when it comes to AI transformation is their mindset. I don’t have the time. I’m not technical. I’m scared to admit that I don’t know how this thing works and ask for help.
I’ve heard it all. But I have also seen retirees and senior executives build agents faster than recent computer science grads. And I have seen twenty-something journalists ship more code in Claude than venture capitalists investing in AI and SaaS tools.
Whether someone becomes AI-native or not is, in my experience, determined primarily by their intrinsic motivation and willingness to tolerate discomfort and learn something new. And if they’re in a professional environment that gives them the space and encouragement to tinker, all the better.
Not all employees will be responsible for or need to be building with AI. But amid a profound labor transition, every single one of us will need to learn how to work with this technology in order to scale our collective impact.
Ibrahim Rashid is a senior associate for impact investing at Sorenson Impact Foundation.
Guest posts on ImpactAlpha represent the opinions of their authors and do not necessarily reflect the views of ImpactAlpha.
