NicholasE — Sagetap Product Pitch Call (Transcript)

Nicholas: (0:00) How are you yesterday?

Bryan: (0:01) We’re good. We’re based here in New York. You’re in Sweden.

Alex: (0:05) Yes, New York.

Bryan: (0:08) Yeah, in New York. What, what city in Sweden are you in?

Nicholas: (0:15) Malmo, southern part of Sweden.

Bryan: (0:17) Yeah, I’ve been to Malmo a couple years ago.

Nicholas: (0:19) Yeah, don’t you have a little cup right now?

Bryan: (0:26) Yeah, yeah, it’s kind of fun and exciting. It’s ridiculously expensive and a pain in the ass to get to the game, so we’re not really, not really, but I’m loving the energy of like all these, you know, European fans coming and discovering random parts of America,

Nicholas: (0:43) Sweden, Monterey in Mexico, discovering that part of the first caves were played on there. So,

Alex: (0:54) yes, Mexico is great,

Nicholas: (0:57) but you guys are playing against what was it facing

Bryan: (1:03) into Bosnia and Herskovinia, I think we’re,

Nicholas: (1:05) yeah, okay, yeah, yeah, they can be pretty good,

Bryan: (1:08) yeah, it’s definitely not the easiest of the second round matchups, the first round matchups,

Nicholas: (1:13) they actually being in Italy, I think in the playoffs in the European close, we’ll see nice prospect. Yeah,

Bryan: (1:26) I was in Stockholm week before last for a couple days. Okay, yeah, different, very different part of Sweden. It’s like talking about like New York and LA or something.

Nicholas: (1:37) We don’t like those people up there. Well, I seem to sell ourselves to be more Danish than Swedish. I know, because you’re right across from Sweden. Yeah, these parts of Sweden actually longer to downward than Sweden, actually. So, yeah,

Bryan: (1:54) interesting. Yeah, the one time I went to Mama, I was in Copenhagen for the week, so

Nicholas: (1:58) yeah,

Alex: (1:59) October.

Nicholas: (1:59) I can almost see Copenhagen from my house.

Bryan: (2:04) Cool, awesome.

Nicholas: (2:05) But perhaps I can start introducing myself. Yeah, my name is Nicholas. I work as EM for, we can call the large retail companies, and in my department we are 200 engineers working in 20 plus teams, and it is a lot of all of our searches are in production today, and the new setup is that people are switching teams to in some kind of idea to get us less silo when it comes to knowledge, so some people jumping around, which it takes a little more effort from our side, but in the long run it will benefit us, so that, and of course, we’re using AI agents in our work to speed up things, and that works pretty well, but this switching between teams, it’s not that easy when we try to pick up things that are amount of people have done, and the same thing also goes before each team had their own on call support, which means that they have to act pretty fast on incidents that are critical, and we’re talking about minutes, and then the developer itself knew what was what they had done right, because there was one who deployed it, but today it’s like a cross-functional team who cover all of all the services to make it more efficient and to save some money, and with that said, they’re not really not have been deployed, and it’s quite hard for them as well to trace it back, especially when it’s AI-generated code, and they have many reasons as well. We were all cases where also we don’t have these md files in our repos that can explain the service itself, so it’s quite hard. And then I saw you guys, you, your service is doing similar to help us, I guess, with these kind of things. And so, yeah, that’s why,

Bryan: (4:18) yeah, no, I’m excited, and I think we’re relatively new to Sagetap. I’ve been building enterprise software companies for about a decade now, so have good amount of experience kind of meeting and exploring and doing discovery and engaging and solutioning all that, but Sagetap is still relatively new. So I believe you applied to the initial campaign I put up back in March, and then I saw you apply it to the new, more product-focused campaign, since, since we, since I kind of began doing discovery in March, we went live with, like, the initial beta of the product, and would love to kind of dive in. That’s why I created a new campaign that’s more of more product-oriented, and I’m just curious, on your end, how have things evolved just in the last couple months, since, yeah, like, for you, like, what are kind of the moving pieces right now

Nicholas: (5:10) in our organization? Faster response time when it comes to production issues is one big thing, also to make this knowledge spread between our engineers more stable, you can call it, because we brought in, we had a lot of our material consultant called independent contractors, who were all of them had to leave, and then we probably want the big consultant for us to replace all these people, and with those people, we’d also be a little bit more, I should call it all over the place, basically to help us to begin a broader view of our organization, because

Alex: (5:57) it’s quite complex,

Nicholas: (5:58) it’s an order management system for our company, so all orders to go through our warehouses, customer customer stores, online shopping, and so on, go through our systems, and it’s 1 million orders per day going through our systems, and so, yeah, we need to have a good picture of that working, because it’s a lot of microservices connected today. Yeah, kind of hard to grasp it. That’s why we try to spread the knowledge, but also with your in production with everything, so we need to be able to step in when we have had some issues when people on call can’t really deal with the issue, because they don’t know about,

Alex: (6:42) yeah, yeah. How much of the code that is going into production now was written by, like, with agentic tools?

Nicholas: (6:51) I think all more or less.

Alex: (6:53) Okay,

Nicholas: (6:53) I can’t say it’s all into its degree, but they’ll more than all engineers are using some kind of tool. Go ahead.

Alex: (7:01) Yeah, I mean, that’s very relevant for us,

Nicholas: (7:04) yeah. So it should be on, so we encourage people to use it, but so yeah, I would say 90% perhaps.

Bryan: (7:16) I think you mentioned in the application, or you know, like the question answering on stage tab, that some of the workarounds that your company, or like your teams, are using, like sharing notes and PRs, like Jira comments, copying, pasting prompts into Slack, is that still the sort of best best practice?

Nicholas: (7:39) In well, not say it’s a best practice, but at least some teams are still doing it, and it’s also one issue, however, because this is all these teams are doing in their own way.

Bryan: (7:52) Yeah,

Nicholas: (7:52) we have not really aligned on we’re working around this yet, so it’s quite often you have to text, they come up with some ID and say, okay, we do like this, okay, and everyone falls all along, and then we have 18 different ways of doing things, so, but yeah, I would say most of the teams that is still relevant,

Bryan: (8:16) and so would it be valuable? Actually, related question, you mentioned a lot of third party consultants that you work with, are they also committing code

Nicholas: (8:26) AR,

Bryan: (8:28) and I imagine it’s even harder to understand them, or to go back to them and ask them to explain the things they’ve worked on,

Nicholas: (8:36) it is, and there’s also a cultural aspect we are located here in Europe, and mostly in Sweden, and in the Netherlands. They are located in India, and even though this is software engineering, it’s not really the same. I can, I can see by CR where do people come from?

Bryan: (9:01) Yeah,

Nicholas: (9:02) how it’s written. So, yeah, but they’re committing, definitely. But sometimes you need to ask, why is this?

Bryan: (9:14) So, would it, would it be helpful to have a requirement enforce that any code that is provided by third parties that they have to provide the full agent session history of the like the work behind the work

Alex: (9:30) quite good,

Bryan: (9:33) okay, that’s those are the bones of what we built to launch with obviously the the enforcement and you know requiring them is that you know, as a contractual thing with you and your consultants, but Alex is going to show a demo now. If we have time, I mean, since we still have 20 minutes left, it’ll be a pretty quick demo. You’re welcome to self demo, right? Like, this is all ready to use today. Okay, so I’ll say, like, the messaging on this site path base is like initial, the landing page is specifically around code review and the PR process, and it’s not fundamentally limited to that. I think that’s just sort of like the messaging that we’re leaning into. So, with that caveat, I’ll have thoughts.

Alex: (10:14) Yeah, the general shape of the product right now, and we’re currently working on adding like general team support, which I think would be really valuable for you, but the shape, as it is launched, like v zero of it is

Bryan: (10:24) as an

Alex: (10:24) individual, you can kind of upload all of your sessions to what we call repositories that are just like all of the individual sessions in a set, you can organize them, and we’re adding tags and things, and then for an individual session, this is actually like a cloud session, we support clod and cursor and codecs, and pretty much all of the major ones, and if there’s one we don’t support, we can add it pretty straightforwardly. This is our general view with attribution of like token spend per prompt, and the general shape of these things. You can see tool calls the whole nine yards of all of this with this view giving a basic like breakdown on where input tokens and output tokens were spent, what the general like since then desk like general latency on requests to the back end cumulative tokens, but we also just added full breakdown per tool and file type and general context of where time was spent on any of this time tokens spend general calls, you can like jump to the part of the conversation that has that, and we’re currently working on adding like query APIs for you to be able to say like give me all the circumstances that this thing occurred, show me any time this tool failed or when this many tokens were spent on a response, those types of questions, does this kind of like make sense? Do you have any questions about this?

Nicholas: (11:48) No, it makes sense.

Alex: (11:51) Yeah, so then the general idea is for the team’s product, we’re working on adding functionality for being able to see this over time for your whole team and be able to have it consistent for all your whole history, so imagine being able to at the moment. How this is constructed, you can resume an existing session in any provider, so you can imagine one of your Indian contractors using Cloud Code or something, uploading their session, linking it with the PR, and then any of your engineers can resume that locally and ask questions of it to be able to directly get to, like, well, what were the decisions made associated with this? Why were they made, etc. Longer term, we’re interested in deriving data from this. What does it look like for people to be able to, like, what were the decisions made associated with all of these features released? Why were they made that way, having those query able by agents, so that you don’t have to like fully resume the PR and everything.

Bryan: (12:45) Yeah,

Nicholas: (12:47) okay, okay. The

Bryan: (12:50) fundamental idea is that the agent session should be a dynamic, reusable artifact, and that if Engineer A works through something, Engineer B, who’s either reviewing it or wants to pick up their work, should be able to essentially pick up that session, and the should not be restricted to what harness or model they use. So that’s that’s the idea, is that engineer A could say work in cloud code, engineer B could boot up the same session when they’re reviewing the code from engineer A, but do it in Codex or Gemini or Pi, or the harness of their choice with the full session history, and then interrogate the agent and ask questions, you know, why did you make this change here to speed up the process of understanding why changes were done, and you can use that today. Yeah, that’s totally ready

Nicholas: (13:35) today, and it’s like a diary, almost.

Bryan: (13:38) Yeah, yeah, I’ve used different analogies, it’s kind of like taking a cadaver out of the morgue, and you know, turning it on and asking who shot you, or, but yeah, I think this is this is all like the baseline functionality that is available for anyone to use for free today. Yeah, this is this is live production that I’m showing.

Nicholas: (13:58) Okay, it’s almost like the engineer itself, who wrote the first initial artist, can actually go back and answer this question themselves, because I think that can be quite relevant as well. Yeah,

Alex: (14:12) I use it, so these are like a bunch that I have to load it for testing and various things. I need to actually name them. These are the like auto, auto named when I don’t give a session a specific name, but it’s actually been pretty useful for sharing on PRs, for like looking back myself, of like how did my tokens like it’s it’s amazing to see how many tokens you actually spent across time as well, and what and where those were. I didn’t realize this session took 124 million cash tokens, like seeing the shape of it, and seeing there’s some interesting.. I auto collapse the like intermediate things, because often I have multi day sessions, and you can see like where I had one day and then it passed over to the next, and what did I do between those? And I normally collapse those, but you can see like breakdown of where time was spent on various things. How often did the agent wait for me? When did it get stuck, what were the contexts there? useful feedback for people.

Nicholas: (15:06) Yeah, I think that’s good, because I think it’s quite easy to get lost in these this AI tooling right now. You can only get a little bit lazy as an engineer, then you stress give too much trust, basically. And then you just commit them. Yep, you’re done, so this is good. I think I’m quite helpful, so but then you have to log into this path base, right, to get this overview. Or how does it is for each commit, or is it for each repo? How does it lower?

Alex: (15:40) So, right now, how it works at the moment is you upload the session at any point in time that you want, and then, okay, this is available, you can upload it anonymously if you want to just do that, but that’s like a temporary period, and I’ll eventually be cleared, if you have an account, we’ll retain it forever, for the time being, you can, without

Bryan: (15:58) signing up for an account, you can create functions like a GitHub gist, right? That’s just like a unique obfuscated link, and that link can be used. Anyone that has that link could then basically pull down that session and hydrate it in the agent of their choice. Yep.

Nicholas: (16:14) Okay. Okay.

Bryan: (16:15) Do you want to show, maybe like from the command line? Sure,

Nicholas: (16:18) I can. Yeah, please. Yeah, let me open

Alex: (16:20) a new..

Bryan: (16:35) It’s always fun when you’re actively building all day, and the terminal in the episode, they kind of switch context to giving a demo that never seems to go away. Yeah, so the

Alex: (16:42) general shape here is we have this path command, and you just do path share, and it’ll actually pop open and show me all of my sessions recently, and I can see a preview on top, and I pick a session. Let’s just pick this kind of medium size session, and it’ll upload it right now. It’s running locally. This is running against my local instance, but it’s uploaded, and then I can share you that session. Let’s see, where is.. I think that’s this one. Yeah, and this is the session that was uploaded. It’s just that easy. And then, if I want to resume it, let me switch back, and just to set

Bryan: (17:20) context, because I was like a lot of screen jumping right here. What happened when he hit path share is the raw logs from clock code were sort of transpiled into this agent agnostic session trajectory format, and then uploaded to the site and indexed, and you know, all these sort of like analytics and metrics are generated automatically, and so when you copy that link and paste in the browser, that was a freshly uploaded session. So this all happens, you know,

Nicholas: (17:49) okay,

Bryan: (17:50) I, you know, don’t have like SLAs, you have to like, you know, sub second, generally

Alex: (17:53) for generally quite fast. And then when someone wants to resume it, they just run path resume, and then the link, and this will work on product. Just happens to be I had a local development session. In this case, it’s resuming it in my home directory, whatever. It’s fine. And this is the actual session that I just ran. And if you had created the session, it would be just as quick for me to download and resume it without having to have a whole bunch of extra effort.

Bryan: (18:15) It

Alex: (18:16) works with all of the harnesses that we’ve spent a lot of time on. The general dynamic that’s interesting, is every harness has its own back end for all these things work, and we do all the work for you to make it so that it’s one consistent way to share and resume across any coding agent.

Bryan: (18:30) Yeah, and if you had this path CLI command installed at the moment, you could, and again, in this case, he’s working off those local pathways, but if he shared a link from one of the path-based hosted sessions, and put it in the Zoom chat. You could copy-paste that and run it, and actually boot up his agent on your machine. Yep.

Nicholas: (18:51) Okay. Okay.

Alex: (18:53) Yeah. I mean, last week you actually used it to switch from Claude to Codex, because Claude was having trouble with something. Yeah,

Bryan: (18:58) I mean, Claude goes down once a week, right? And it’s like you don’t want your work to suddenly stop, and so I’ll just use it locally without even going up to Pathways, and just basically hand off the session from my local cloud to my local codecs and continue on. It’s actually kind of a funny thing, is like you make this switch and you know Codex isn’t aware of this, right? Codex just boots up and assumes that it did all this work because it looks like it didn’t work. It’s actually very funny, and like, I made a.. so I started a cloud session, and I said, “You know, change our website to be purple, right? And it just turned for a few minutes, and you know, suddenly the website was purple. And then I did path resume and switched over to Codex, and I was like, “Codex, why did you do this? Like, well, you know, I wanted to make the website purple because I thought it looked better. It’s like, it just.. it has no, it doesn’t really know, it just, it all it has is its history, and it uses that to kind of base all of its, you know, intelligence,

Alex: (19:48) yeah,

Bryan: (19:50) but yeah, so there’s a lot of like handy engineering flows,

Nicholas: (19:55) yeah, Matthew, I can really see purpose with this, right, so also in your.. is this your website? Path-based, yeah, path-based dot

Bryan: (20:06) dev is like the product landing page for the for this service, and empathic dot dev is our sort of general, you know, startup landing page.

Nicholas: (20:16) Okay, yeah. And I guess it’s pretty easy to start using it, right? Yeah,

Bryan: (20:20) yeah,

Alex: (20:22) yeah, and at the moment it’s, it’s like focused single users, what we released, and right now I’m currently adding like team support and ACLs and everything,

Nicholas: (20:31) but what I would like to do is I will take this back to some engineers in my teams, yeah, haven’t tried out and you see what you feel about it, then we can perhaps have a follow-up call after that, see how we can,

Bryan: (20:51) yeah, and if any of your engineers want to jump on a call to kind of like go through the initial steps, if they have any questions, I’m happy to make, you know, make a make us available for that, yeah. Of

Nicholas: (21:02) course, of course, I just want to see who’s working, because I’m based in Sweden, and you’re in July, no one is working mid summer, yeah, yeah, you shall see each other again in mid August, so yeah, but I think there are one or two people that should be otherwise services will be not. So

Bryan: (21:24) I believe you said it’s over 100,000 people at your company,

Alex: (21:29) over 200,000 people.

Bryan: (21:32) Okay, who else like you there is forward thinking and trying to get out ahead of this? Is there anyone else that might be interested in if you’re trying to build some consensus internally,

Nicholas: (21:41) engineering department is quite large. I think you’re over 5000 people working in engineering, so but I cannot surround my colleagues just to get an understanding. I learned to mention a company that worked on Francis, we’re based in Sweden.

Bryan: (22:05) Okay, cool.

Nicholas: (22:07) So, you can have a guess.

Bryan: (22:10) Awesome. Yeah, no, I’m.. this has been great to connect, and looking forward to getting, yeah, getting feedback. I think, of course, yeah, I think we’re both looking for individual users just to battle hard in the platform, get feedback from engineers, make sure we’re staying aligned with the needs of what people are actually doing and put in the ground, also looking for design partners, some sort of immersive two week thing, maybe to kind of dive in and help integrate with their teams and help improve workflows and things like that,

Nicholas: (22:41) that sounds good,

Bryan: (22:42) cool. good, cool,

Nicholas: (22:43) great.

Bryan: (22:45) Okay, awesome.

Nicholas: (22:45) Thank you, guys, and have a really great

Alex: (22:54) thank you would be a good customer

Transcribed by https://otter.ai