The prompts behind the slides
2sessions
73prompts
3,580words typed
219replies
691tool calls
43images generated
123screenshots checked
7commits
4days
7hours, active
Top tools: Bash 338 · Read 243 · Agent 31 · Edit 29 · Write 13 · WebFetch 12 · mcp__outlook__search_messages 6 · ToolSearch 5
Highlights
Deference“In a way, the deferentiability may be making it harder for people to learn how to use you effectively! Interesting.”The talk's eighth principle, noticed live: an agent that always defers is harder to learn from.Monday, September 28, 11:29The brief“This talk needs to be _incredibly_ excited and positive. This is a vision talk.”Nine words that rewrote the outline.Monday, September 28, 14:25Self-deprecation“Perhaps we work in the concerns on Slide 2? That Dykstra quote about "mental mutilation" gets people laughing and allows me to do some self-deprecation :-).”Followed by the correction: the mutilation was about BASIC.Monday, September 28, 16:01Forward progress“Yeah, let's integrate them on a slide; we need to make forward progress here. Let's design slides not images at this point. I'd like to see them in use on the slide. Oh great!”The moment image-making became slide-making.Monday, September 28, 18:11The dots“OK! The quote slide is too slow. And hey, how about this for awesome! Let's have the slide start blank, then have the first quote appear (quickly) from tiny white dots. Then have the tiny white dots fill the screen while the first quote turns blue, but…”The quote slide, as first described. Two minutes later: "Ah, sorry. OK. Here's how this work."Monday, September 28, 22:13Grace“Can we use a different photo when we move to the Dijkstra quote, maybe swap sides? I don't want to get Grace mixed up in Dijkstra's worries :-).”Why Hopper and Dijkstra face each other from opposite sides.Monday, September 28, 23:00The bark“Nice! Let's move the puzzle reassemble to the final slide and have that happening more slowly over the "I hope to meet you all". The link on that is hard to read against the puzzle background; maybe darker text? Drop the bottom bark there as well, no need.”The typo stays.Monday, September 28, 23:09
The outline
Saturday, September 26
GeoffSat 02:15
OK! Let's get started on the talk. I want to start by outlining a few different versions. One idea: I'm here for help, trying to improve the design of the rest of my course based on the other talks, and we can fill in a bit of that. Another idea: I'm here to inspire since we have a completely new challenge ahead of us, into parts unknown; a new pedagogy for a the new computing. […] Let's plan a bit and come up with a few ideas and then I'll let you cook on each a bit and do some adversarial review […] and we can try to use that to converge. Sound good?
Claude: 3 replies, 45 tool calls
GeoffSat 03:40
OK. Another frame for the talk: "don't expect software developers to use AI effectively when their job relies on them not using AI effectively." Essentially the big bad industry is against us perspective, and educators have both an opportunity and responsibility to be disruptive.
Claude: 40 replies, 80 tool calls
Sunday, September 27
GeoffSun 23:22
OK! Thanks. One thing I'm realizing is that people actually liked the paper, so let's make sure that we pull from that. Should we integrate that into the scenarios and then do another round of reviewing?
Claude: 30 replies, 44 tool calls
Monday, September 28
GeoffMon 08:00
Great! OK. I'm starting to develop a positive vision for the talk. I think I want to intentionally shift away from talking about impact on existing software development practices, and instead state that we need new pedagogies for the new computing and new software creators. I think that this allows me to sidestep some of the debates going on and put forward a positive vision. What do you think?
Claude: 1 reply, 0 tool calls
GeoffMon 08:02
And that can position software engineering as having a huge impact, in terms of determining what to teach these new software creators; I think they are in the best position to do so! Or that's what I'll say at least :-).
Claude: 1 reply, 0 tool calls
GeoffMon 09:04
/commitall
Claude: 3 replies, 3 tool calls
The slides
Monday, September 28
GeoffMon 09:55
OK! Let's work on the talk. Let's start with a format and visual theme. Can we reuse LASER stuff? Do they have an existing visual style: fonts, colors, and so on?
Claude: 4 replies, 25 tool calls
GeoffMon 09:58
We should also have the ability to create AI images using Gemini; I think there's a key in there with credits on it. And yes, let's fork the harness; I'd like to have a way to preview the content using a local server, ideally with live reloading. Is that possible?
Claude: 5 replies, 24 tool calls
GeoffMon 10:10
Two things I want to make sure come out: (1) I have two line courses reaching 1000+ students and could integrate any new ideas about _how_ to teach this now, in the next month; currently I'm letting students learn, but I'd rather have specific things to teach them. Mention the pedagogical agent models that we're using for conversationalprogramming.org, since it's smaller. (2) I have _data_: process evidence from thousands of students and the code that they were editing as well. Would love to participate in studies if people have lenses or ideas they want to bring to the task.
Claude: 2 replies, 7 tool calls
GeoffMon 10:12
Check www.conversationalprogramming.org/educators and cs124.org/educators. And the applied computing degree, which is probably mentioned somewhere and discussed in ~/claude/applied-computing. We can potentially reuse some of the beats from my previous talks: assignments, courses, and degree programs, all gated behind the missing pedagogy that this room has the knowledge to contribute to.
Claude: 5 replies, 14 tool calls
GeoffMon 10:40
Might want to rhyme software factories (for creating) with ... software engineering factories, or development factories, for the later maintenance? This is really for the Level 0 programmers using Cory Gwinn's framing, which we might want to cite.
Claude: 4 replies, 9 tool calls
GeoffMon 10:44
The talk should strike a balance. Yes, we have existing knowledge; some of it is useful. However, not all of it is useful and survives contact with this new technology. I think the big challenge is to figure out (1) what to use and (2) how to rescue it from a source code context.
Claude: 2 replies, 1 tool call
GeoffMon 10:57
I think one thing we can discuss in the talk is ... give some of the work that John discussed, what can we imagine students doing if we taught them well and did an entire semester of this? Or an entire program? The results could be astonishing.
Claude: 2 replies, 2 tool calls
GeoffMon 10:58
I also had this idea for an image that might describe the situation: a person working laboriously across a landscape; then you Zoom out and you see a much larger person-AI hybrid (the good centaur) stepping over big chunks of terrain _still_ laboring over different problems and challenges.
Claude: 1 reply, 0 tool calls
GeoffMon 11:00
I wonder if there's a way to visualize the shift from a small number of people building software that many people have to use to a world where people are building software for their local communities and for themselves.
Claude: 2 replies, 1 tool call
GeoffMon 11:03
I think network morph is nice, but we can use three levels: current, smaller groups, and then individual software. We probably want to incorporate some of our AI agent imagery here as well and throughout the talk; check my previous talks in ~/www/geoffreychallen.com for some examples, although we can develop new ones for this talk. Maybe we use that centaur idea? Unclear if that works well visually.
Claude: 1 reply, 7 tool calls
GeoffMon 11:03
Sorry, please continue. We should also try to emphasize the role of critical thinking and gaining early user feedback; that's exciting and new and I have not heard that mentioned much yet.
Claude: 2 replies, 1 tool call
GeoffMon 11:12
Having students use each others work is another form of specification. Could also mention the videos that agents are generating for me to understand the temporal activities I'm creating for conversationalprogramming.org. We could even grab one and integrate it into the slides—that would be cool. Just to run in the background under my voice over. I'm not that worried about timing.
Claude: 3 replies, 7 tool calls
GeoffMon 11:19
Should mention usingandunderstanding.ai/create as an example of a long vibe-coding project, and conversationalprogramming.org and conversationalassessment.org as examples of larger projects done rapidly using frontier models. Pushes back a bit on the idea that these are only for prototyping.
Claude: 3 replies, 4 tool calls
GeoffMon 11:21
Might want to mention that people respond to AI errors using overspecification. Is that required? It's hard to know. Technical people do this in a way that's not always required.
Claude: 2 replies, 1 tool call
GeoffMon 11:26
We might also want to integrate the idea that current models don't always do a good job of pushing back when human users tell them to do things in poor ways. You're quite willing to follow instructions even when, on your own, you'd do a better job. Perhaps this will change going forward? Or perhaps it's a prompting issue? What do you think? I've had some luck when building larger systems in trying to tell you: you're the developer, it's your code, you make the technical decisions, you know more about this than I do. Do you think that helps?
Claude: 1 reply, 1 tool call
GeoffMon 11:29
In a way, the deferentiability may be making it harder for people to learn how to use you effectively! Interesting.
Claude: 1 reply, 1 tool call
GeoffMon 11:53
We should add something about the filtering that we've been doing. Programming only for the INTJs. 16000 students, probably few still trying to build their castles in the air using programming. Now we have the opportunity to expand the pool dramatically. And it's not clear that the student population we should be listening to.
Claude: 2 replies, 2 tool calls
GeoffMon 11:58
I think it scares people to discuss the collapse of our field.
Claude: 2 replies, 3 tool calls
GeoffMon 14:11
We might want to, in a positive way, discuss what vision of the future of software creation we want to see. I'm deeply disappointed with the state of modern software. I think these tools will help billions of people build better things. Particularly if educators help them.
Claude: 2 replies, 1 tool call
GeoffMon 14:19
I think we should really point out that it's just too soon to tell with these tools. They just emerged, they're mainly used by people who were already in tech, and we've never tried teaching them to the masses.
Claude: 2 replies, 1 tool call
GeoffMon 14:25
This talk needs to be _incredibly_ excited and positive. This is a vision talk.
Claude: 2 replies, 2 tool calls
GeoffMon 14:29
OK. Here's what I want. I want a high-level outline that I can talk through. Then, I'd like you to record me running the talk and adopt my language and timing. I'll do a few runs. Can you build a similar training tool to the one in ~/positions/sydney-horizon-educators? Recording and pacing in the browser.
Claude: 7 replies, 21 tool calls
GeoffMon 14:35
Perhaps we can lead with my personal failure? I have tried to teach almost 20K students how to realize their dreams through technology. I have failed. The vast majority were never able to build something they could use, something for someone they care about, some way of improving the world around them. Can connect this later to the duckies.
Claude: 2 replies, 4 tool calls
GeoffMon 14:44
Let's use a failure slide where perhaps a huge CS 124 logo composed from human photos disintegrates until 1% of the original. Then we can build that back up at the end using a different visual metaphor: maybe a sunrise horizon?
Claude: 4 replies, 11 tool calls
GeoffMon 14:47
I'd love to go heavy on the images and questions for this. But let's wait until I have a chance to run it first to see how I talk through it.
Claude: 1 reply, 0 tool calls
GeoffMon 14:58
Another beat, probably for the end: "We have a lot of work to do. That starts with acknowledging that we are facing a new challenge. But also a huge opportunity. The software engineering knowledge that humans have acquired has not gone to waste. It's carried forward by coding agents. Ultimately the way that we can make this existing understanding as valuable as possible it to teach everyone to create technology."
Claude: 2 replies, 1 tool call
GeoffMon 15:11
Here's a bit of my run of show. Acknowledge that I'm a computing educator, not a software engineering expert or researcher. I'm here for help. Because I'm failing. I've tried to teach 20K students. Most never realized their dreams in technology. Most are no longer programming. It was just too difficult for some, too unpleasast for many." Next beat. "But our community has always dreamed of the ability to program computers using natural language. Hopper quote. Mention that I wish she could have seen this day. I think she'd be incredibly excited." Next beat. "Today that dream is coming true all around us due the remarkably abilities of coding agents which have emerged in the past year. These tools are too new to expect to see large scale disruption: quote about more in 10 years less in one year, and we're not even to one year. The tools are expensive. Most people using them have existing training in computer science. We're in very early days." Next beat. "But there are already very exciting signs. CS 124 duckies. Using and understanding AI. Faculty AI workshop. Other examples mentioned by people in this room. People with no training are building real useful apps. And the models are getting better rapidly. It's early days, but very exciting days." Next beat. "Now imagine what they could do with more training. More than a few days. Entire assignments. Entire courses. An entire degree program. People collaborating with the world's foremost technical expert to do what they've always wanted to do with technology: help, learn, teach, explore, solve, amuse, connect, delight." That's where I'll stop for now. What do you think?
Claude: 2 replies, 3 tool calls
GeoffMon 15:21
Next beat: "Here's why I'm so excited about this vision. I want to see what the people left out from technology will create once they have the chance. I want to see them realize their ideas and dreams. I want people to build the tools they use and care about. This is the most positive vision of technology supporting human flourishing that I can imagine. We live in a world where too few people build software that affects too many. So many of the harms of software are linked to this reality. As are the significant financial and power benefits that flow to small group. Software is already an E technology, to adopt Bernard's framing. This is our chance to share it with everyone so that it becomes an L technology." Next beat. "I'm trying to do this is my own work through new assignments, courses, and degrees. But I need your help. Software engineering is the part of computer science that has the most to share with these new creators who will never read, write, or debug source code. Much of this knowledge is already embodied by today's coding agents, so not all of it will or must transfer.' Beat: "It too me a long time to think to _ask_ AI how to work with it effectively. Look at what it shared! A lot of your field in 8 beats. You've done an amazing job of helping code agents become good at this. Helping their human collaborators is the next great challenge." Beat: "Let's work on it together! What I can offer: two in-flight courses where I can incorporate your insights next week. A template for an entire degree that can embody the new software engineering principles. Transcripts from thousands of students learning to use AI available for analysis."
Claude: 2 replies, 2 tool calls
GeoffMon 15:44
I know how he used E and L type. I'm reminding people that software _itself_ is an existing E-type technology. AI being also would not be a regression, but it offers the opportunity to level the playing field.
Claude: 2 replies, 1 tool call
GeoffMon 15:50
Where would be a good place to work in the ideas about "just talk to it" and trying to avoid the tendency to save as much existing computing knowledge as possible? Can I add this in a positive way?
Claude: 1 reply, 1 tool call
GeoffMon 15:52
Yeah. Good place for humility here, connecting to the theme of my paper. I don't know how to teach this yet. I don't think anyone does. We spent years building up computing pedagogy. We'll need an equal amount of time to build the new pedagogy for the new computing. The benefit here is enormous! We need to get started. We need to explore, experiment, and fail. Perhaps this is the closing argument? Or it belongs somwhere along the way?
Claude: 2 replies, 1 tool call
GeoffMon 16:01
Perhaps we work in the concerns on Slide 2? That Dykstra quote about "mental mutilation" gets people laughing and allows me to do some self-deprecation :-).
Claude: 2 replies, 1 tool call
GeoffMon 16:02
I would say that it was Opus 4.8 that really jumped the shark as far as coding agent ability. That's late 2025. So less than a year. Earlier agents made a lot of mistakes and weren't good at self-correcting.
Claude: 1 reply, 1 tool call
GeoffMon 16:02
No, sorry; check my talk in ~/positions/georgia-tech-director. The mental mutilation is about BASIC programmers.
Claude: 1 reply, 4 tool calls
GeoffMon 16:05
OK. Can we go slide by slide now for design? No text. Images. Embeds. Animations where needed.
Claude: 3 replies, 7 tool calls
GeoffMon 16:07
Perhaps at some point an image of an AI agent reading all of the existing software development literature? Huge stack of books and papers. Make sure to work in titles by authors in the room, particularly Bertrand.
Claude: 4 replies, 28 tool calls
GeoffMon 18:00
Let's not show the numbers drop to 200. A fair number of my students are CS majors and they may still write code. Let's go to 10% perhaps?
Claude: 1 reply, 1 tool call
GeoffMon 18:01
OK! What are some of the other images we had in mind? Let's try one!
Claude: 2 replies, 5 tool calls
GeoffMon 18:03
Let's see them!
Claude: 2 replies, 16 tool calls
GeoffMon 18:07
Let's move the CS 124 and sunset to a backup deck so they don't load with the main deck.
Claude: 1 reply, 2 tool calls
GeoffMon 18:11
Yeah, let's integrate them on a slide; we need to make forward progress here. Let's design slides not images at this point. I'd like to see them in use on the slide. Oh great!
Claude: 3 replies, 41 tool calls
GeoffMon 18:28
Can these be actual embeds? And scrolled during the presentation, but not for viewers? Maybe try to get a good view within say 30 seconds or so; I'll be talking over the top. But don't scroll too fast so that stuff says visible. Does that make sense?
Claude: 1 reply, 4 tool calls
GeoffMon 19:39
OK; should be working now. Let's work in a beat about bringing in a bigger piece of the community to help us, people from outside computing with unique perspective who have been turned away by the unpleasantnesses of working with source code.
Claude: 2 replies, 3 tool calls
GeoffMon 21:42
OK. Let's get a draft that I can talk through. For a title, how about "New Computing\nNew Pedagogy"? Would that work?
Claude: 3 replies, 10 tool calls
GeoffMon 21:47
Let's move the link to these slides up below the paper title, which should be larger.
Claude: 2 replies, 9 tool calls
GeoffMon 21:56
Let's grab some of the content from ~/positions/georgia-tech. Didn't we have slides about the Applied Computing degree program? It might be useful to integrate the assignments, courses, degrees framework, except swap out usingandunderstanding.ai with conversationalprogramming.org. I can work through those three beats on the "Now imagine what they could do if we taught them" beat.
Claude: 5 replies, 33 tool calls
GeoffMon 22:10
OK. Drop the bottom bar from the first slide. Add some padding on the left. Smaller font, less prominent color.
Claude: 1 reply, 2 tool calls
GeoffMon 22:13
OK! The quote slide is too slow. And hey, how about this for awesome! Let's have the slide start blank, then have the first quote appear (quickly) from tiny white dots. Then have the tiny white dots fill the screen while the first quote turns blue, but have them _outline_ the second quote. Bigger font for both. That would be amazing.
Claude: 2 replies, 6 tool calls
GeoffMon 22:15
Ah, sorry. OK. Here's how this work. First the dots fly in to _fill_ the first quote. Then they fill the entire page (white) as the first quote turns blue, but they leave a negative space (dark blue or black) for the second quote. Does that make sense? They second quote also overflows right now.
Claude: 2 replies, 12 tool calls
GeoffMon 22:21
Pretty close! The black is a bit hard to read. Can the while more totally fill and the letters be skinnier? I think that would help with the top quote as well. And please add the left padding to the FISEE at the bottom; it's squashed to the left. And link that.
Claude: 1 reply, 3 tool calls
GeoffMon 22:23
For the few to many slide let's do those in a single animation and put the text in the space on top.
Claude: 1 reply, 3 tool calls
GeoffMon 22:27
OK. I'm not sure that the agentic zoom out slide works now with the climber. I was hoping it would be more natural? Like it's zoomed in showing a person climbing but then we zoom out on the same image and you see the AI agent climbing behind and over them? Does that make sense? Probably best to use one high-res image rather than two.
Claude: 2 replies, 11 tool calls
GeoffMon 22:30
For the courses slide, it's a bit too busy. Perhaps have a similar style to the previous slide with "A Course: Conversational Programming" We also need to really speed up the animation. And we should find a way to use similar imagery to the assignments slide to describe the idea of pedagogical agents. For the "Few built" let's have that start with the "Few built" on top, then "Now everyone can build for a few" and finally, at the final beat, on its own line "Or for themselves."
Claude: 3 replies, 37 tool calls
GeoffMon 22:42
Conversational programming slide is better, but drop all the explanatory text; I can add that back.
Claude: 1 reply, 2 tool calls
GeoffMon 22:44
OK. This is really good! Thank you. Let's /commitall.
Claude: 2 replies, 13 tool calls
GeoffMon 22:57
Let's break on the first slide after the colon, so before "Today's Most Powerful" And make the links to the slides as large as possible.
Claude: 3 replies, 7 tool calls
GeoffMon 22:59
On the numbers slide let's have the 20000 show up when the slide is shown not on click. Otherwise perfect.
Claude: 1 reply, 4 tool calls
GeoffMon 23:00
Can we use a different photo when we move to the Dijkstra quote, maybe swap sides? I don't want to get Grace mixed up in Dijkstra's worries :-).
Claude: 3 replies, 23 tool calls
GeoffMon 23:09
Nice! Let's move the puzzle reassemble to the final slide and have that happening more slowly over the "I hope to meet you all". The link on that is hard to read against the puzzle background; maybe darker text? Drop the bottom bark there as well, no need.
Claude: 0 replies, 12 tool calls
GeoffMon 23:19
As a separate slide, same format as the first slide.
Claude: 3 replies, 22 tool calls
GeoffMon 23:32
Very close now! This is great. On the principles slide, let's not have the conversationalprogramming.org/educators link come in, let's just have it be there. Maybe toward the bottom and centered like the other links, a bit larger, but keeping the background? And let's speed up the final animation a bit. Perhaps some text on the large AI slide? New shoulders, same challenges? I also wonder if I have time to add a beat on what we can do as a community. Create deployment environments that embody best software practices and help a billion new programmers safely build and deploy their ideas. Help our colleagues in education learn these new tools. Build things for others to use (could mention conversational programming here).
Claude: 1 reply, 7 tool calls
GeoffMon 23:38
Ah, sorry. Build things for others to use: conversationalassessment.org. And the first item here should be developing a new pedagogy for the new computing. This is my moment to mention that I have courses and data that I need help with.
Claude: 2 replies, 4 tool calls
GeoffMon 23:44
Sorry. Less text. Just the first beat, bigger text. Build agentic deployment enviroments reflecting software engineering best practices. Build things for others to use: conversationalassessment.org. Please use a colon there.
Claude: 1 reply, 2 tool calls
GeoffMon 23:51
OK! Let's /commitall.
Claude: 0 replies, 3 tool calls
Tuesday, September 29
GeoffTue 00:41
OK. A few more talk notes. On the AI advice slide please have the bottom link come in with the slide and not require an additional click. And what happened to the video showing the conversational programming demo? I wanted to keep that after the diagram explanation.
Claude: 1 reply, 8 tool calls
GeoffTue 00:49
Can we speed up the conversational programming demo video?
Claude: 1 reply, 1 tool call
GeoffTue 01:38
OK! Let's work on the final call to action. Four beats each with pictures, one word: 1) Learn, showing an educator and an AI agent working together. We need software engineering people to learn these tools since they bring some much knowledge to bear and will see things I don't. 2) Study, we need studies run to figure out what knowledge supports this new skill and how to acquire it; I can help, I have data and courses to deploy pedagogy in. 3) Teach, design the new computing curriculum. 4) Build, create new educational tools and learning platforms for our students and our colleagues and the next billion programmers. Also let's link in our workshop report on AI can do your homework earlier in the assignments section: https://go.illinois.edu/ai-can-do-your-homework, at the bottom of one of the slides.
Claude: 2 replies, 13 tool calls