May 2025

“These aren’t just chatbots. They’re designed dependency machines,”

The Deep View did a nice job back in mid-May staring emotional companion chatbots right in the face.

https://www.thedeepview.co/p/the

Companion bots highlights two equally important realities, IMHO,

1. Humans are still highly social creatures that require social and emotional activities to remain healthy.
2. If you are making a chatbot, bring a social psychologist onboard to ensure your app does no harm.

Number 1 is easy to wrap our heads around.

Number 2 provides an opportunity to prevent unintentional myopic thinking, echo chambers, and IRL isolationist behavior (e.g. opting to only engage with chatbots rather than live humans).

Why is this important?

Myopic thinking and isolation fosters “othering” behavior, which leads to hate and bias. These things left unchecked have never proven to be a good thing.

So, if you are building chatbots, use your powers wisely.

tl;dr: The rise of AI Agent Development is reshaping how we develop software, create professional development pathways, and teach coding.

A recent job post from Firecrawl caught my eye: they’re offering $5,000/month to “hire” an AI Agent—built by someone else—to autonomously perform content creation tasks (Firecrawl Job Post).

I’ve been thinking a lot about this whole “people who build AI Agents” thing, and this job post validates it: a new software stack is emerging, and it’s the AI Agent Development. We’re entering an era where AI Development is becoming a distinct discipline—sitting right alongside Front-end, Back-end, and Database development.

Firecrawl is leaning into this future by treating AI agents as first-class contributors—products in their own right, not just tools. They’re not only contracting these agents, but also providing a potential pathway for their human creators to join the company full-time.

It’s a bold signal: the people who build and orchestrate AI systems are becoming central to modern tech teams.

This shift changes everything. Professional development (PD) for developers will need to evolve, accounting for developing, integrating, collaborating with, and managing autonomous AI systems. The same goes for how we teach programming—skills like prompt design, agent chaining, and orchestration will become table stakes.

Firecrawl’s post is a glimpse into a future that’s already here.

I was reminded of Dr. Buolamwini’s words when I read about a study finding gender bias emerging from some LLMs. The tl;dr of the study:

• If queries presented female (i.e. name, language, or impression), responses were simplified or redirected to less technical stuff (e.g. design over coding).
• If queries presented male, responses includes more detailed steps and technical language (e.g. jargon).
• Bonus-Ugh: If queries presented female, responses were 23% more likely to include phrases like “Don’t worry if this seems complicated.” Aka, the LLM assumed a more emotional response from the user.

Other evidence of our past in the data: Jobs are gender stereotyped.

”Women (were) mainly assigned job titles such as graphic designer, fashion designer, or nurse and men assigned job titles such as software engineer, architect, and executive… ChatGPT has a hard time associating male pronouns with nurses and an even harder time letting female pronouns handle a pilot’s duties of getting a plane ready for landing.”

Why Is This?
Our collective unconscious lives (and thrives) in the training data. Genders are associated with certain jobs because the source material does so. The associated responses to technical questions is because of the training data.

What Can Be Done?
All three of the below are critical to improve our collective training data:

1. Be aware of what you put online. Remember, LLMs often use web content to train on.
2. Report what you find. Most LLMs have a reporting mechanism. Use it.
3. Support an organization.  Some organizations include:
   1. Distributed AI Research Institute (DAIR)
   2. Center for Responsible AI at NYU
   3. Montreal AI Ethics Institute

References:
Buolamwini, J. (2023). Unmasking AI: My journey to hold AI Accountable. Penguin Random House.

Emelyanov, A., & Chuprina, S. (2025). Ethical and security aspects of multimodal foundation models. Array, 19, 100295. https://doi.org/10.1016/j.array.2025.100295

Kennedy, P. (2024, March 22). New study finds gender stereotypes persist in ChatGPT. TechXplore. https://techxplore.com/news/2024-03-gender-stereotypes-chatgpt.html