The Intern Test: A Mental Model for AI Readiness in Your Workplace
Would ten new summer interns boost your team's productivity or hurt it?
Picture this: It's a bustling Tuesday morning at the office. You're knee-deep in your work when an excited HR person bounds up to your desk. "Great news!" they exclaim, eyes sparkling. "We had an amazing showing at the college fair. You've got ten new interns starting this summer!"
Does your heart leap with excitement at the prospect of fresh talent and extra hands? Or do you feel a sinking sensation in your stomach, thinking about the time and effort that might derail your team's productivity this summer?
Your gut reaction to this scenario might reveal more than you think. In fact, it could be a telling indicator of your team's readiness to integrate AI into your workflow. Let's assume, for argument's sake, that AI agents capable of enhancing your team and automating mundane tasks are real and on their way. Would it actually help your team?
In this article, we'll use the mental model of the "Intern Test" to investigate which roles and teams have workflows that can effectively handle task delegation and supervision to make the most of AI automation.
The Supervision Paradox: Why AI is Like an Intern
One of the hurdles in adopting generative AI in the workplace is the risk of hallucinations - instances where AI produces plausible but incorrect information. While efforts to mitigate this issue are ongoing, we currently must accept some level of "making stuff up" as an inherent characteristic of these systems. This reality makes thinking of AI as an intern a useful mental model. Just as you wouldn't assign an intern to interface with a major client without proper training and supervision, AI requires oversight.
The key to effectively integrating AI lies then in identifying tasks that are hard to do but easy to check. Why is this particular combination is ideal? let's take a look at a 2x2 matrix of task difficulty and verification ease:
Easy to do, hard to check: In this scenario, using AI offers little benefit. The extra work required to verify the AI's output outweighs the simplicity of the task itself.
Hard to do, hard to check: There's a threshold at which AI becomes useful here, but it requires careful consideration. For instance, AI can assist radiologists in reading reports, but the model must meet a high bar for avoiding false negatives. This need for human verification is why we haven't achieved full autonomy in fields like aviation.
Easy to do, easy to check: These tasks are typically handled efficiently by humans already but AI can provide scale here.
Hard to do, easy to check: This is the sweet spot for AI integration. It's where we're seeing the growth of AI-assisted tools like GitHub Copilot. In these cases, AI provides suggestions and updates that humans can quickly verify, offering significant productivity gains.
This matrix illustrates why the "hard to do, easy to check" combination is ideal for both AI and interns. It allows for meaningful contributions while making sure there is supervision. In the next section, we'll look at workflows that are particularly well-suited for both interns and AI, shedding light on the characteristics that make certain tasks ideal for delegation to these "supervised assistants."
Are Your Tasks Chunkable and Independent?
One of the key indicators that your workflow is ripe for AI integration is the ability to decompose work into independent, verifiable subprojects. This characteristic not only makes it easier to incorporate interns into useful projects but also paves the way for effective AI adoption.
Consider content creation workflows, where AI has gained significant traction. Marketing organizations have long been accustomed to hiring freelance writers on platforms like Upwork or Fiverr to produce content they later edit and publish. This existing workflow made it natural for companies like Jasper to offer AI-generated content at a fraction of the cost. Instead of paying $150-$300 per article, companies could access AI writing tools for a flat monthly fee, changing the economics of content production.
This model works well because the work product is a completely independent chunk. If an AI-generated article doesn't meet standards, you can discard or revise it without disrupting other processes. The ease of verification and the independence of the task make it an ideal candidate for AI assistance.
Companies that are already used to hiring and firing contractors are well-positioned to adopt AI solutions. These organizations have developed processes for decomposing work, farming it out, and assimilating it back in after review. While there may be additional editorial and management work required when using AI, it's largely a matter of scaling an existing workflow.
This "chunkable and independent" characteristic isn't limited to content creation. We see similar patterns in graphic design, where companies might use AI tools to generate initial concepts or variations, then have human designers refine and finalize the work.
The key is that these tasks can be clearly defined, executed independently, and easily verified without disrupting the larger workflow. As AI capabilities continue to expand, we're likely to see more industries identify areas where work can be broken down into these AI-friendly chunks.
Does Your Workflow Have Tooling to Reliably Integrate, Test, and Revert Work Product?
While some industries benefit from tasks that are easily chunkable and independent, many workflows involve interconnected processes where team members build upon each other's work. In these scenarios, the ability to seamlessly integrate, test, and if necessary, revert changes is important. This is where robust tooling and established processes play a vital role in AI readiness.
Consider software development. To produce complex systems with advanced functionality, it typically requires teams of developers working in concert, building on each other's contributions with millions of lines of code. Teamwork is so integral to software development that source code management systems are nearly as old as modern software itself.
These systems, like Git (popularized by platforms such as GitHub and GitLab), offer essential functionality:
Version control: Tracking changes over time and maintaining a history of revisions.
Branching and merging: Allowing developers to work on features in isolation and then integrate them back into the main codebase.
Collaboration tools: Facilitating code reviews, issue tracking, and project management.
Reversion capabilities: Enabling teams to roll back to previous versions if new changes introduce problems.
This infrastructure allows development teams to add new members - be they human interns or AI - with a degree of confidence. However, the integration isn't always smooth. Having worked with distributed teams, I can attest firsthand to the importance of fit. An intern (or AI) slinging bad code can slow down implementation and create problems. Even with crisp software requirements and robust testing and validation processes, concerns about security issues and code quality persist.
The good news is that the software development industry has already cultivated a culture of integrating work from various contributors, including those who might not be known personally.
Other industries are following suit. In design, tools like Figma allow for collaborative work with version control, making it easier to integrate AI-generated designs into existing workflows. In content creation, content management systems (CMS) like WordPress or more advanced platforms like Contentful provide mechanisms for staging, reviewing, and publishing content, which can accommodate AI-generated material.
The key questions to ask about your workflow are:
Do you have systems in place to track changes and revisions?
Can you easily integrate work from multiple contributors?
Is there a process for reviewing and validating contributions?
Can you quickly revert changes if there are problems?
If you can answer yes to these questions, your workflow is likely well-positioned. The ability to reliably integrate, test, and revert work product provides a safety net that allows for experimentation with interns or AI while maintaining the integrity of your processes.
Do you form new teams for projects?
Some industries assemble teams from their employees for projects, often including new members who haven't worked together before. These environments, while lacking the structured infrastructure of software development, still demonstrate a readiness for AI integration due to their processes around project management.
Consider fields like law, consulting, and investment banking. In these professions, teams often "swarm" around large projects such as mergers and acquisitions, fundraising efforts, or major litigation cases. These projects require the collaboration of many different specialists, each contributing their expertise to the final work product.
For instance, in an M&A deal, various team members might need to review documents in a deal room to extract crucial information. Similarly, in large-scale litigation, junior associates in law firms are accustomed to spending hours sifting through discovery documents, highlighting important ones for the partners on the case. These tasks, which require attention to detail and the ability to extract relevant information from large volumes of data, are candidates for AI assistance.
Automation tools have already proven their worth in these scenarios. AI can efficiently process large volumes of documents, flagging relevant information and potentially uncovering insights that might be missed by human reviewers. This type of task aligns well with the "intern test" - it's challenging work that can be easily verified by more experienced team members.
However, when it comes to contributing to the final work product, the bar is significantly higher. You wouldn't want an AI system citing a non-existent legal precedent or sending the team on a wild goose chase with an invalid legal argument. This caution is well-founded, as recent research has shown.
A study by researchers from Stanford RegLab and the Institute for Human-Centered AI highlights the pervasiveness of legal hallucinations in state-of-the-art language models. The study found that hallucination rates range from 69% to 88% in response to specific legal queries. Moreover, these models often lack self-awareness about their errors and tend to reinforce incorrect legal assumptions and beliefs. These findings raise significant concerns about the reliability of Large Language Models (LLMs) in legal contexts, underscoring the importance of careful, supervised integration of AI technologies into legal practice.
Some questions to consider for this aspect of AI readiness are:
Is your team accustomed to incorporating new members quickly for specific projects?
Do you have clear processes for verifying and validating work from team members?
Can you identify lower-stakes tasks where AI assistance could be valuable?
Do you have experienced team members who can oversee and verify AI-generated work?
The Real Gatekeeper: Your Team's Culture and Processes
Our journey through various team structures and workflows has revealed the obvious: the readiness for AI integration - or even effective use of interns - hinges on a team's ability to decompose projects, incorporate new contributors, and maintain quality. These requirements might seem like Project Management 101, but their implementation often proves more challenging than it appears on paper.
Consider the assembly of a car or the assembly of a board deck. In theory, these projects can be neatly divided into clear sub-tasks: engine, frame, and transmission for the former; marketing, development, and G&A slides for the latter. It sounds intuitive, doesn't it? Yet, anyone who's worked in large teams knows that reality often diverges from this ideal.
In many organizations:
Requirements exist more as vague notions than well-documented specifications.
Decisions are made by HiPPO (Highest Paid Person's Opinion) rather than data-driven consensus.
Tribal knowledge about metrics and workflow consequences is scattered, often decided in ad-hoc meetings without key stakeholders.
This dysfunction isn't just a minor inconvenience; it's a barrier to improvement and automation. It's why management consultants command hefty fees for re-engineering companies and teams. McKinsey & Company's research suggests that 70% of change programs fail to achieve their goals, largely due to employee resistance and lack of management support. Similarly, a study by Gallup found that only 13% of employees strongly agree that their leadership communicates effectively with the rest of the organization.
Some industries and functions have been forced to confront these challenges head-on. Software developers, for instance, have had to build robust systems to manage bug fixes and code updates, leading to the development of sophisticated version control and collaboration tools. Consulting firms have mastered the art of rapidly assembling effective teams for new projects, creating standardized onboarding and knowledge-sharing processes. These industries have evolved out of necessity, driven by the high costs of errors and the need for flexibility. However, many other sectors haven't faced the same pressures, leaving them ill-equipped to integrate new contributors effectively, be they interns or AI tools.
In such an environment, even the most advanced AI tools or the brightest interns will struggle to make a meaningful impact. The smartest algorithm can't overcome unclear objectives, and the most eager intern can't navigate a maze of undocumented tribal knowledge.
In essence, the "Intern Test" we've explored throughout this article is as much a test of our organizational readiness as it is of AI's capabilities.
As we move forward, the most successful teams won't just be those with access to the latest AI tools. They'll be the ones who have done the hard work of creating a culture and infrastructure that can effectively leverage these tools - and any other innovations that come their way.
Is your team ready?