Consider receiving an an occasional nudge.

We might get these sporadically and indirectly in a work environment. You’re “thinking” or browsing the web, a college wonders over, you talk, it shake you lose and you get back to work.

So, we can bot-ify this, but better?

Drop in the daily TODO’s into doc, have an LLM check the doc and monitor your screen or project directory. They can then decide how and when to give you a little nudge to “get back to it” or “push beyond”.

An office assistant, but not clippy. Not annoying.

Motivating.

Inspirational.

Helpful.

Personalized.

Optimized.

It could pose questions directly related to what you intend to do and what you have done to push you in new/different directions, e.g.:

  • “did you consider this or that”
  • “what are your thoughts on this or that”
  • “did you see this paper related to this thing we’re working on”

Then a/b test intervention strategies, time of day, etc.

A nudge-bot or, with alteration: boost bot.

From our gpt friend:

A boost is an intervention that actively enhances a person’s competence, skills, or decision-making abilities so they can make better choices independently. Unlike traditional nudges, which subtly steer behavior, boosts provide knowledge, training, or tools to help people achieve their goals.

Yes, a boost, not a nudge. Nudge is too passive. We are making active interventions here with goal of more/better output from the target.

I’m sure the kids are working on versions of this, but it’s not hard to MVP this if armed with some Python and some APIs.

Real-Time Real-Life Boosts

There was a time in the late 2000s when people started fact checking on Wikipedia in public, e.g. at the pub or at a party.

We want from “people talking crap” to “fact checking people’s crap”.

I noticed it in the next crop of phd students coming through our lab.

I think it happened because Wiki was getting really good/useful around that time. And we all had or were starting to get smart/iphones that could hit the web.

Now we can LLM-fact check stuff.

We do with the kids all the time. Or a topic turns into an ad hoc trivia Q&A session with a gpt bot.

So fun!

There are tens of tiny things we could be checking-in with or resolving with a boost bot during the day.

  • “How can I get the dog over the back fence to stop barking”
  • “Should I trim these things into a hedge?”
  • “Is eating this a good idea for me?”
  • and on.

Hmm, not a boost bot….

Like a siri that is not annoying.

I guess this is what home automation thingo’s like alexa offer.

A “Thinking Partner”

Hmm, so a writer boost-bot…

The default is auto-spell and grammar checks. The bot is monitoring the text and suggests things to make the language correct.

Okay, so grammarly is doing this and more.

We can imagine all manner of syntax/semantic suggestions, structure suggestions, correctness suggestions, etc. The expected stuff.

What else? Things like:

  • Encouragement that we’re onto something interesting
  • Suggestions of where to take it and ideas to include
  • URLs for things to link to that support our points
  • Concepts from the literature that give names to the things we are writing about
  • People who’ve had similar ideas and what they said

Build it into the CMS (wordpress) or text editor as a plugin. Lame.

Na, built around this idea from the ground up. Like “paragraph comments” and “wiki talk page”, but real-time by our helper agent community.

Don’t write for me, help me think better while I write.

A thinking partner, again, from our gpt friend (emph. mine):

A Thinking Partner Agent is an AI assistant that guides your thought process while writing—not by writing for you, but by helping you think more deeply, clearly, and critically about your ideas. This agent would act as a collaborative interlocutor, asking strategic questions, prompting reflection, and encouraging creativity and depth in your writing. It functions like a thought-provoking mentor or Socratic dialogue partner, enhancing your cognitive engagement with your work.

Yep, “collaborative interlocutor” exactly.

I do a version of “this” now, but I have them (my agent round table) open in separate tabs, and it’s manual.

Scratching my own itch :)

From gpt4o:

Core Principles for a Thought-Enhancing Agent

  1. Conversational & Responsive – Feels like a dialogue, not just a tool.
  2. Cognitively Engaging – Encourages deeper thinking rather than just correcting.
  3. Non-Interruptive but Available – Offers insights only when useful.
  4. Customizable & Adaptive – Learns what types of thinking prompts work best for you.
  5. Encourages Metacognition – Helps you reflect on your thought process.

Yes: “Non-Interruptive but Available” and “Encourages Metacognition”. These things together.

Think about “paragraph comments”. They help because they are context specific. Think about “reviews”. The help you by offering alternate perspectives.

I think tools like sudowrite and novelcrafter and probably doing stuff like this, although boxed into specific subtasks for fiction writing.

I should try it out one day.