Professional Development – 2026 – Week 42

To write a good goal, you need to figure out all of the small roadblocks the agent my encounter. Give the agent tools it needs so it doesn’t have to ask you.
You need clear exit criteria that are based on high-quality evaluations.
You will need to spend more time up front planning the work, finding good exit criteria, and providing tools and skills for self-evaluation and self-correction.

This TED talk described three types of revelation (the inspiring act of revealing something).

Wonder walk — talk based on revealing a series of images, wonder built over a series of steps (often used by artists, designers, photographers, and architects). You need a clear through-line to link all pieces together with a goal to give the audience an inside scoop of your experience in clear language.
Dynamic demo — show the audience something in action. Have something impressive to demonstrate. Tease, explain, demo, then show the impact.
Dreamscape — share a vision to inspire others (e.g., screenwriters, inventors, architects). Use image, demos, sketches, words. Paint a bold vision of the future and make your audience desire that future.

It’s not X, it’s Y. Choppy declarative list of sentences. Phony revelation that ends with a question.
LLMs pick words that are technically correct but sound a bit off.
Seeing formulaic writing throughout a piece that’s similar to writing elsewhere that follows the same formula.
Awkward linguistic flourishes and segues that don’t really work. This feels like a high-school student padding out a statement to increase word count.
“Gets that part right”, “this actually matters”, “honestly not a deal breaker”, “here’s what’s interesting”. For example, an LLM might say, “Here’s what’s interesting. Baboons are actually very violent creatures.” A human might say, “The interesting part to me is that baboons are actually pretty violent.” People don’t say things like a scholarly journal, we share our own experiences.
Starting sentences with “because.”

Unintended sophistry = writing that sounds well-reasoned and confident, but doesn’t say much
Structure of a statement, a list of three things, and a summary.
Em-dashes
Regular writing tends to be more opinionated and personal. There’s uncertainty and nuance. Epistemic hedging. LLMs aren’t self-aware that it picked a particular word/token, and hedging doesn’t sound confident (which is it’s default).
To fix:
- Write like you talk and use AI to prune your babble to compress them down to the most important ideas.
- Ask LLMs to critique your work rather than rewriting it.
- Read it out loud and see if it sounds right.

One failure mode of mob/ensemble programming is where one person is typing and everyone else is passively watching (or is checked out). Having an agent write code is another version of this.
The second failure mode is where everyone is talking and the typist is confused. This also happens once an agent finishes writing some code and everyone is picking apart what the agent did and what to do next.
Quiet prompting = individuals come up with a prompt, then all are revealed, and the group chooses the best one (or combine/mix)
See: https://lexler.github.io/augmented-coding-patterns/pattern-catalog/
Have one person read (out loud) what the agent did so you can understand it together and decide whether it makes sense or what changes are needed.
Keep 15 minutes at the end to have a retro on how the ensemble programming session went.
This is collaborative engineering using agentic tools.
Note: Could be good for mentoring junior engineers, or helping people incorporate AI.

Geoff Mazeroff