Closing the Year with Reflection and AI Insights
Thoughts on growth, teaching, and two reports shaping how we think about AI.
Happy Holidays
This is my last post of the year. I will be away for a few weeks, spending time with family and friends. It will be a time to think, reflect and define my goals for next year, including putting together a list of New Year resolutions.
Reflections on This Year
Before shifting into self-reflection mode, I want to pause and celebrate a few accomplishments. Launching this newsletter has been one of the most exciting milestones for me. While I am not focused on driving high traction, I use this space to document my journey and share what I learn. My first post was published in February, and I am proud to have stayed consistent throughout the year.
I also ran several AI workshops for finance professionals and moderated a panel discussion for the first time at SF Data Summit. I became an instructor at WorldQuant University, which has been a rewarding experience that allowed me to teach and interact with students around the world.
On the personal front, I used AI to assist with kindergarten research. While some argue that statistics do not fully capture the true value of a school, AI helped me consolidate reports so I could identify a few schools worth visiting. I have toured nearly fifteen schools, which helped me understand the education landscape in San Francisco. We are not applying this year and are simply gathering information since I did not grow up here.
Anthropic Interviewer Findings
To close out the year, I want to share a few interesting reports. Anthropic introduced an AI interviewer earlier this year to help researchers gather data and analyze user behavior across the economy. The initial dataset includes 1,250 interviews across three groups: the general workforce, creatives and scientists. The data is publicly available.
The interviewer works in three stages: planning, interviewing and analysis. At the planning stage, a system prompt is created that includes hypotheses about each sample and best practices for designing the interview plan. The interviewer then generates specific questions aligned with the research goal. Human researchers review and refine the plan before moving forward.
During the interview stage, the AI conducts real-time, adaptive interviews. In the final stage, human researchers work with the AI to analyze transcripts.
The results highlight a mix of optimism and anxiety across all groups. Eighty-six percent of professionals in the general workforce reported that AI saves them time, and most described AI as augmentative. Creatives primarily use AI for administrative tasks and remain cautious about peer judgment. Scientists use AI for literature review, coding and writing, but many report frustration due to hallucinations and the extra time needed to verify results.
A separate analysis by Dejan, an AI SEO agency, summarized the findings well:
• General workforce: AI is treated like an eager junior intern who handles smaller tasks with supervision.
• Creatives: AI streamlines administrative work, while creators retain full control of the core creative process. Anxiety about peer judgment and job security is still common.
• Scientists: AI behaves like a super librarian with a lying problem, offering helpful information but requiring verification.
Perplexity Research on Agent Adoption
Another interesting study comes from Perplexity in collaboration with Harvard researchers. By analyzing interactions from Comet and Comet Assistant, they uncovered patterns around agent adoption and usage.
While many still rely on AI for routine tasks, more activity is shifting toward cognitive work. Agents are becoming thinking partners rather than simple task executors. Fifty-seven percent of agent activity involves cognitive tasks, and thirty-six percent relates to productivity and workflow improvements. Another twenty-one percent centers on learning and research.
For example, a finance professional might delegate tasks such as filtering stock options or analyzing investment information. The agent gathers information, summarizes insights and, when requested, helps implement decisions.
The research also shows a clear pattern of users starting with low-stakes queries. Once they successfully use AI for complex work, such as analyzing financial statements, the dependency becomes sticky with strong retention.
Closing Thoughts
These two reports reinforce a future where humans and AI collaborate closely. I have mentioned this concept many times, and it is encouraging to see broader consensus forming around the idea of AI with a human in the loop.
This feels like the right way to conclude the year, knowing that AI applications are becoming more mature across different industries and use cases.
Happy holidays, and I will see you next year.