Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
How Engineers Can Use AI Right Now: Five Workflows Worth Trying!
36 Plays3 days ago
Join Jason and Carlos as they unpack another three topics from the world of construction:
🚂 Canada's High-Speed Rail Ambitions: Deep dive into the proposed Toronto-Quebec City connection, examining how the project shifted from high-frequency to high-speed designation, the staggering $50-100 billion price tag, and why a journey that was faster in 1970 now needs such massive investment.
🤖 Practical AI Applications in Construction: Live demonstrations of five game-changing AI implementations anyone can use today - from automating concrete pour analysis and cost tracking to streamlining subcontractor pricing negotiations, and building an automated early warning system using site WhatsApp messages.
😊 Construction Happiness Insights: Analysis of the first Construction Happiness Barometer survey of 300 professionals, revealing why colleague relationships trump salary for job satisfaction, and what this means for project leadership and team retention.
Key Timestamps:
00:00 - Introduction
06:02 - Canadian High-Speed Rail Project
12:07 - AI Uses In Construction Right Now
37:32 - Construction Happiness Barometer
Check out the Off Site newsletter here
Follow The Offsite Podcast for weekly insights into the biggest stories in construction and infrastructure.
Recommended
Transcript
Introduction and Hosts
00:00:00
Speaker
ah cool project hopefully it gets going even if it does there's a lot of design timeline and probably another kind of example of you know in the same time it takes to make this decision and design the thing China's probably built you know 400 times more rail. Probably about a couple of miles while we've been speaking. Yeah
00:00:27
Speaker
Welcome back to Offsite Podcast. I am Jason Lansini and joined again by Carlos Cavallo. I'm speaking from Germany. Carlos, you are in London. How ah is your day going, mate?
Personal Anecdotes and Topics Overview
00:00:41
Speaker
Day's going okay. um Although I have a baby who decided to not sleep last night, so if I'm flat, it's ah it's not my fault. It's the disclaimer.
00:00:51
Speaker
I am on coffee number four and a coke with caffeine. I'm in the same boat. so um so um We have a bunch going on in the news and a bunch of things to cover today. So I'll just quickly run through what we're going to talk through on the docket. So first of all, we're going to look at the very briefly the Canadian high speed slash high frequency slash the pivot from high speed to high frequency or the other way around ah rail project that is recently in the news.
00:01:26
Speaker
Then we're going to show each other some projects. We took some homework and some projects away to build some form of ah how we would implement AI as a person on site today. So we're going to do some show and tell of those projects.
Humorous Stories and Cultural Comparisons
00:01:44
Speaker
And then our last on the list is the latest construction happiness barometer.
00:01:50
Speaker
um So spoiler alert, we're not real happy and we'll talk about that and why. You're settling into the European mindset really well. You come from happy Australia and you're already...
00:02:03
Speaker
No, i don't know I don't know about happy. We're definitely wearing less clothes.
High-Speed Rail Projects in Canada
00:02:09
Speaker
ah But yes, in the news this week, ah the number one thing that stood out to me, which above all else in a very crowded news cycle is some beavers built a dam. Talk to me about what's going on.
00:02:24
Speaker
So in some very questionable news websites, and I guess a shout out to Emplan's Alan Mosker for sharing the first on LinkedIn that we picked up, there's a town in um and the Czech Republic which have been debating, arguing, and trying to secure a million quid for a long time to build a dam. And then in just a few days, Beavers decided to do it for them, obviously tired of waiting.
00:02:50
Speaker
so the the article was poking fun at the fact that all of this red tape bureaucracy and struggling to find budget a few beavers did in a low carbon super fast way sorry the beavers found budget or they found it they i don't they want it but uh they've actually managed to do it um ending the requirement for a dam so they're obviously they've already done some sort of cat three check on this dam and decided it's a permanent structure ah That's not the right engineering time, is it? Cat 3 check? Is that it? I don't know. he did You mean beaver 3 check? you yeah um me check no the yeah So the headline, it was too juicy not to jump at, which was like in the Czech Republic, Beavers built a dam and saved authorities $1.2 million. dollars And seven years of construction. Seven.
00:03:42
Speaker
You're right, it's a very suspect website, but I saw it on a post from Alan and if nothing else, he's trustworthy and would have hi would have definitely not shared something that was not highly cited. Yeah, but said enough said. um All right, so let's dive into Canada's high frequency rail. So highlight of the highlight of the project is effectively, it is a proposed connection between Toronto and Quebec City.
00:04:12
Speaker
It is one of the biggest rail projects that Canada has ever proposed to undertake. It's proposing to replace an existing corridor or service that ah effectively is notoriously slow and unreliable ah because basically you have commuter and freight rail on the same network.
00:04:33
Speaker
um and the kind of newsworthy thing that came out recently is that it was originally proposed as ah high frequency, whatever that means, and then has recently kind of decided to supersize and added the idea of high speed and high frequency. Interesting project Carlos, what are your thoughts on, well what are your thoughts on what high flow what high frequency, high speed means ah in this context. So it looks like the trains are going to do 200 kilometers an hour, which apparently in and the club of high speed rail networks, it's not technically high speed, so they revalue high frequency. I thought I saw 300 because 300 is definitely up there, right?
00:05:16
Speaker
Three hundreds up there, but the current design is two hundred. So this news in October must be around. Let's make it real sort of high speed. we Let's go 50% faster. Yeah. Yeah. So like China, France, Japan, they all have over three hundred kilometers an hour networks. So it's not quite high speed, which is why the name is odd.
00:05:35
Speaker
And then on your on your comment around unreliable trains, ah they're aiming for 95% on time, um because currently the round network in Canada is 70% on time, which is actually pretty poor stat. The three trains are delayed. That's like as bad as Germany, which I've recently found out is very bad.
00:05:56
Speaker
Yeah, exactly. But I guess just ah to tap in a little bit around what they're actually building for anyone who is unfamiliar, um it's kind of ah the western corridor of major cities um being connected. So Toronto, Ottawa, Montreal, and Quebec, which is about I think straight line distance 850 kilometers, I think it's a thousand with the including the the class or the actual line of the track. Apparently, it was quicker to make that journey by a train in 1970 than it is now. thats So that's a pretty poor stat. It's 55 years later. That's like construction productivity. Yeah. he has um and it And it very much looks like they're spending all of their money serving one, what is actually a tiny slice of the country. So if you check out ah that route and zoom out on maps, it's actually just a tiny section of Canada right in the West.
00:06:45
Speaker
and neglecting the rest of the country, an astronomical price where they're touting 50 to 100 billion and potentially. Yeah, exactly. Yeah. It wouldn't be the first country to spend all their money on highly densely populated centers and ignore the kind of regions. and Yeah. um But it's the first intercity rail line to be built since the 80s.
00:07:08
Speaker
Since then, China's built 42,000 kilometers of high speed rail. So um they're very much behind the curve of connecting these cities. Yeah, as you said, 50 to a potentially even more than 100 billion. ah The article was published um on the offsite newsletter, ah which people should definitely subscribe to. Also notes that there's over 150 daily flights between Montreal and Toronto. So that is a lot of people moving.
00:07:37
Speaker
Yeah, it's it's going to be a real
AI in Construction: Current and Future Applications
00:07:39
Speaker
behavior and almost sort of um habit shift because between these those major cities just in that side of the country, only 5% are made by train at the moment. So even if you have that train service, breaking into people's commute and routines, I think would actually be really difficult.
00:07:58
Speaker
when they're so used to jumping in the car because I can imagine sounds like something I could have looked into before but I'd imagine it's not like a highly interconnected city in most of these cases where you can get to one of these major stations really quickly you can imagine a lot of driving a lot of sort of spread out urban areas or suburban areas so yeah Before everyone packs up and moves to ah Toronto in order to deliver this project, it's worth noting that the design process is supposed to last at least five years ah and that will cover at least one election cycle and it would be
00:08:38
Speaker
On form for the project to potentially get cancelled, there was a previous high speed rail project in Ontario and announced in 2017 that was then abandoned in 2022. So hopefully not. I wouldn't say pack your bags and fly there.
00:08:54
Speaker
right away. um But yeah, there's every chance with these big politics, you know, these big infrastructure mega projects that politics ends up trumping. Well, then on the topic of trumping, there's also the the risk of all the tariffs that have been ah so recently announced. And someone needs to find, ah yeah, $100 billion dollars is going to be potentially harder to find for a government that is now facing pretty extensive tariffs. Yeah, yeah let's not dig into Trump because we'll have to rerecord six times before this is released based on recent information. So cool project, hopefully it gets going, even if it does, there's a lot of design timeline and probably another kind of example of, you know, in the same time it takes to make this decision and design the thing, China's probably built, you know, 400 times more
00:09:46
Speaker
ra probably built a couple of miles while we've been speaking. Yeah. So before you move on, I just want to note that our producer put a comment in chat saying, by the way, 300 kilometers an hour is 180 miles an hour. I think we're on the fly, on the fly translation.
00:10:05
Speaker
All right, let's get into the second topic on the docket, which is AI for construction. Now, this is probably the most over-discussed topic. I feel like we've covered it 600 times ah in different forms in this conversation. But one of the things that seems to be a giant lack of is there's like a lot of hype. There's a lot of talking about what's going to happen and what's going to come. And there's some SaaS applications that are kind of making some noise and running pilots and starting to get some momentum. but
00:10:36
Speaker
For most projects on the ground, their experience of AI is often Microsoft's terrible co-pilot, which is not really helping anyone do anything. you can there's a whole bunch of There's a whole bunch of news out there. Not really helping anyone do anything. What a lie.
00:10:56
Speaker
There's a whole sub-genre there of like how bad that is. um So you put that aside and the experience of most people on construction projects is very limited to you know what it will actually do. It largely looks like hype.
00:11:13
Speaker
So we thought we'd give ourselves the challenge and imagine that we're on construction projects um today and therefore had you know limited tool set to what we could probably access as just a ah normal person that didn't want to spend more than a couple of dollars on anything and try to stick together some examples that we would kind of showcase to each other of how we'd use construction use AI in construction today for kind of the best impact.
00:11:41
Speaker
Now we should preface this with, first of all, um construction company IT t teams ah would hate this idea. This is shadow IT. This is shadow AI. This is people signing up and using the data in AI systems. But yeah, I think at some point, construction IT teams need to start allowing some degree of experimentation here because There's, as we hopefully will show, there's some pretty compelling use cases right now for someone on ah on a construction project. So with that being said, Carlos, do you want to do scissors, paper, rock or also go first? I've never had someone say scissors, paper, rock before, but we can do rock. That's good because I did i said did it. I said scissors, paper, rock. I couldn't even say because I've never heard that phrase before.
00:12:34
Speaker
That's just five copies in a coke that's doing that. i'm happy You can go first if you like or I'm happy to. You're cool. How many how many different ideas have you got? Two. I've got endless ideas. I've got two that I'm prepared to share. I'll start with my like most simple and I'll maybe ratchet up from there. Imagine I'm a construction site engineer.
00:12:59
Speaker
One of the things that I used to do a ton was get all sorts of dockets of concrete dockets that pile up in an in-tray in my desk and then I'd have to enter them into a concrete register that was like all the different concrete pores that had happened on the project. And then I'd get, I'd use that register to do reconciliations when I got an invoice from a supplier. um I would use that to you know, get help do my forecast for the month to work out how much we'd spent, how much I'd reforecast going forward. And a couple of the big time wasters were obviously manually entering all the concrete docket records into into a spreadsheet. And then also just chopping and changing that to do, you know, build some Excel charts and stuff to work out spend um over time.
00:13:50
Speaker
The current state of play is some people are doing this now with docketing systems and they're getting those records coming into a docking system, but I see lots of projects still doing some form of manual record of this.
00:14:01
Speaker
ah So what I did is I signed up for a Claude account or aka in this case, I already had one and then I went and scanned in all the Concrete dockets that I had and then I asked Claude to basically extract a certain number of properties across each of those dockets Which was the docket number the volume where it was delivered the um mixed type and a couple of other properties like that. And then it just iterated through all of those dockets and put that into a markdown file. ah Let me share it with you. Yeah, so I then um so if you look at this one here, I basically uploaded the
00:14:51
Speaker
um PDF documents and then it generated, it basically put that into a markdown file or a TXT file which extracted all the concrete pores with their volumes. And it was at about this point that I ran out of um requests on my keyboard account. As I had to do a quick switcheroo over here to chat GPT.
00:15:18
Speaker
And I basically asked it to, if I can upload that data, I basically copy and pasted it all in um and can help me do some analysis on it. So I did some very simple things to start with. And I only had about 30 concrete pores in in this data set. um But you could easily do this with hundreds of concrete pores.
00:15:38
Speaker
um So at a top level it started breaking down the volumes by different mix types for me. um Whether they had ice in them or not. i assume I was trying to basically break it down by what was a ah cost variable for me.
00:15:54
Speaker
So if I scroll down further, what I then asked it to do is essentially uncover, okay, here are my three different mixes that I've poured in the last whatever period the data set is for. Here's the cost per cubic meter of each of those mixes. I can easily add in the cost for additional ice, et cetera. And I said, basically, just give me a month by month spend breakdown of what it should be.
00:16:22
Speaker
um So I've taken all the PDFs, uploaded them and extracted all the the data from it and then iterated through that data and said basically help me get a cost comparison of what I should have spent um over those periods.
00:16:39
Speaker
And it's basically giving me a cost per month, September, yeah, so August, September, October, and it's breaking down all the different concrete pours, the volumes, the rate of that mix, and then basically the total that I should have spent on concrete. So I could use that to sense check what I'm getting invoiced each month from say a concrete supplier. That job would have taken me, that was normally a Saturday job that would have taken me half an hour to an hour to go through once a month.
00:17:08
Speaker
and I could do it in probably 10 minutes. Massive time save immediately. And probably more accurate. Probably. Yeah, not wanting to malign a 24 year old me, but um yeah, I'd say possibly more accurate. Very good. And I think it's worth just noting for anyone that hasn't experimented with these tools yet.
00:17:28
Speaker
When you're getting these responses and then you're probing with more questions and asking for tweaks, it is updating the memory. so In the future, when you ask similar questions, you're getting immediately to the output that you do want not sort of starting again. so yeah Over time, you save more and more and more. You also personally learn, like once you work out what prompt works and what prompt doesn't work, you immediately like your own mental model of it ah changes and you get straight to the right question to ask yeah in the next run.
00:17:56
Speaker
I'm going to run this one live because ah I'm that confident in my ability is to use this now. ah Nice. I did find myself having a strange conversation with it yesterday and when I was running a rehearsal.
00:18:08
Speaker
ah However, um so i thought i'd I've got two examples. The first one I run through now were really day-to-day exercises where anyone could just pick this up and use it tomorrow. so Really simple.
00:18:23
Speaker
day one using AI, what could I use it for? yeah and A lot of what we're talking about now, yes, IT teams, we're annoyed, but you could really anonymize a lot of this information to still get answers without sharing anything sensitive. so Use that argument with your IT team if you get found out. But um I thought firstly, what a lot of commercial teams or in Australia's case, engineers would do is get responses from market,
00:18:52
Speaker
Pick out some key materials that are the the big cost drivers within those prices and help you do a bit of comparison or analytics on that. Something that you would normally take a bit of quants, add in loads of columns, try and align all the prices and then spend a lot of time getting to what is actually quite a simple sort of response. so I'm starting by saying, hey, I'm running a tender and a prices for key materials for these subcontractors. Can you review and provide detailed analysis to help me prepare for a meeting with each? I'm keen to understand how they compare with each other. Are they market standard rates? So yes, we can benchmark against the responses, but what's market actually saying? And where is my time best spent negotiating?
00:19:37
Speaker
So I'll ask that and it's going to say, yeah, cool. Can I have these prices? um And I'm just going to copy and paste this very simple table. Awesome. So I've just pasted in the data from um just a quick spreadsheet, ah which involves a few subcontractors and a few key materials.
00:19:56
Speaker
it's immediately spit out a bit of um analysis on the price themselves so benchmarking and the subcontractors against each other so this is where you would go through and try and pick out the most favorable rates to sort of leverage the others down. um The second section is actually comparing it to market rates so we can see whether they're within market rates or within a normal bracket and see any outliers so really they want to pick out where things are surprisingly low because you know that it's actually probably not the sustainable, especially in a large contract and in a construction world, they'll probably try and recover costs elsewhere. um And obviously, you don't want to be overpaying for something outside of market rates. And it's immediately given me areas to focus in my negotiation. So with subcontractor to A, this is the rate you should be arguing down, which actually makes their full price quite reasonable um and um it gives you a bit of a crib street to run that meeting.
00:20:50
Speaker
So super quick way of taking some data, um plugging it in and automating a task that would have taken you two or three hours in reality, but it really is just a few minutes. On top of this, so if we go back to the tool that Jason used, um which is Claude. ah Claude is particularly good at pulling in lots of documents. You could go and get things like um the database information for things like BCIS and Turner and Townsend. If you really wanted to give it some in-depth information to then work out on that those rates to see what is market standard down to specific like mixes of concrete or types of reinforcement.
00:21:29
Speaker
So this is a bit of a quick and dirty. You can go super detailed if you've got that information to hand. So I see you were using in that one the 4.0 model um in that example. So that's the one without like reasoning capabilities um in the response. Did you test both? Did you find any benefit one way or the other? I didn't test both, no. I did test ah Google Gemini. I found this one better.
00:21:57
Speaker
Yeah, OK. Well, that probably is a good transition to my next one, and we'll we'll run that live in both examples and evaluate the differences. So the the next use case that I was thinking about is.
00:22:10
Speaker
Everyone knows that the kind of dirty secret of construction records is like 90% of it is in WhatsApp. And that is where basically all of the back and forth about what's happening on site every day and those same projects are struggling and trying to get people to complete site direct records. I thought it wouldn't be good if we could ah get through a whole thread of WhatsApp messages and uncover anything of note in that.
00:22:37
Speaker
So the kind of step-by-step to this ah little trick is that if you're in a WhatsApp thread um that is kind of where people are going backwards and forwards on project records, um ah you know, back and the forwards on, we did this, this happened, you know, we've got this issue, this was delayed, this subby was, you know, understaffed or whatever.
00:23:00
Speaker
um You can go into the thread in WhatsApp and go to the bottom and you can export the group thread basically to a markdown or a TXT bar. You can then create a bot basically. So you can go and create in chatgbt. I'll do it in chatgbt for this example. I would probably preferentially use Claude, but I've hit my token limit for the day.
00:23:22
Speaker
So we're going to do this in chatgbt. I could go to that file, and if I share my screen, I'll run a pretty basic prompt. I'm going to be running this live, so I don't know if it's going to be good or suck. But basically, I just prompted 01. So this is the um reasoning model, which basically will iterate over itself to think through the problem.
00:23:44
Speaker
I generally find that it's like kind of wordy in its responses, basically saying you're an NEC expert, so assume they're in the UK, who analyzes site records, WhatsApp trends, and other construction progress records to identify and draft early warnings. For each document uploaded, review in detail and identify where a potential issue requires an early warning, and then when asked, draft a fully compliant early warning notice.
00:24:06
Speaker
You're representing the general contractor, big things contracting, and it is an NEC4 contract. And then I've just also prompted when identifying early warnings use a table. So if I go to writing early warnings and it's not fun for anyone wondering what that's like.
00:24:23
Speaker
Yeah, I am going to open because chatgbt sucks at broad file support. I just opened my export from WhatsApp in like a text editor and copied the content out. ah So I'll paste that in here and you can see the format. It's like timestamp who said it and it's back and forth between people effectively here.
00:24:51
Speaker
I'll read out for anyone an example. So there's a record from January, Pete Morris, the side super said, yeah, mate, I got two backups now. I'm not taking any chances after that man. So it's kind of like a back and forth thread. So I will then ping that off to it and it's gonna start working through again. This is the reasoning model. So it's basically going over and iterating over potential risks.
00:25:18
Speaker
And remember, I prompted it to kind of put this into a table for me. So here it's now going through, and we'll see how many it's identified. So identify three potential early warmings in a couple of weeks worth of um of WhatsApp threads. So first one is a message about SnapTyrod.
00:25:37
Speaker
um potential risk that other ones could be faulty, drafted it, it's referencing the right clause under NEC4 and the potential impact. So then ah let's go and draft an early one with floor issue one.
00:25:59
Speaker
So that's basically gone and drafted in the structure as identifying the notification, the clause reference where he might give an early warning of a matter which could, and then referencing the three things that you would issue an early warning for, clearly describes the matter and references exactly when it was happened, who said what and when, and then identifying the potential impact and suggested mitigations.
00:26:26
Speaker
You could actually upload the guidance notes from the form of contract that you're in to make sure that the early warning is contract compliant as well. Yeah, totally. you could You could upload the whole contract data. And like one of the things that I would do if I was to put this into production as in really use this on a project today.
00:26:42
Speaker
I'd probably put all of my contract documents. yeah All the works information into a into a prompt and then I would get it to i get that to help write the brief for how we should write early warning so that you don't need all of that information for every time you do this.
00:27:05
Speaker
but you kind of distill what the requirements are for the project. And you can even go further. And even in that example, it's still a tiny lift on your time for such a strong output, creating a project in a cloud, uploading all these documents and it's there and ready to go whenever you need it.
00:27:24
Speaker
Yeah and then like obviously this is a hypothetical early warning notice for anyone that's like seeing my screen but um you could go one step further and in the prompts basically say here's our template of an early warning notice with the headings and the content that you need and it wouldn't need to invent a bunch of the headings in here.
00:27:41
Speaker
Yeah, yeah, for sure. My next one's on a similar theme of writing. um And I was thinking about my time running construction projects and how many method statements I always had to write on a daily basis.
00:27:55
Speaker
so I was actually chatting through a member of our team here who is an exchange engineer to make sure that I'm not way off on this one um because now I haven't written a method statement before called caveat. But I said, hey, I'm looking to draft a method statement. um Here's some information. I'm building a block work wall using the Wembley Innovation System. It has a single door opening at 900 miles wide. Oh, sponsored both.
00:28:20
Speaker
It's four meters high, eight meters long, um with WI columns at maximum three meter intervals. I'm after methodology, risk assessment, and other key insights. So that's all I gave it, and it's very quickly written. A little scope of works, a list of materials and equipment required to build the wall,
00:28:41
Speaker
methodology, broken down into prelims, actually building out your blockwork wall, risk assessment and control measures. and some other sort of key insights or things to consider. yeah um So pretty useful information there. I then said, are three-meter intervals um excessive? yeah And it said, yeah, great question. Actually, it is excessive. Really, you'll be fine for four to five meters without additional reinforcement. So it's immediately picked out something that I could remove, which would obviously... yeah fast in the process. So a real simple one, but you could then upload your template if you really wanted to, and it would write it within the template that you need for the project. but i go yeah The amount of time I spent writing method statements. Yeah, it's yeah or even even just getting a bunch of the ideas or bullet points down or or recording a meeting, talking through it with the team, and then getting it to write the draft. like that It's so such a big time saver.
00:29:40
Speaker
Yeah, and even like control measures to reduce risks like all this stuff you can spend hours thinking about. um Yeah, so um yeah, super strong. Yeah, run a workshop with the team use something like a like a fathom or get a transcript graola or something you can get a transcript from it, feed that in and go, hey, can you go and basically generate a risk register from this? And then you tie that into the draft, I would say the heap of time. Yep, absolutely.
00:30:13
Speaker
All right, we're going to, this is going to be super quick speed run. I went, so these were kind of like basic, just using chat GPT or cord. And then I thought, how far can we kind of push the, the boundary on it? And what could you actually kind of build if you were going to spend, I don't know, a Sunday evening playing around, um, and maybe you wanted to, uh, spend under $5, uh, building a test.
00:30:36
Speaker
And so I took the idea of like early warnings and site records and WhatsApp threads and being able to draft early warnings and thought what would that look like if you kind of wanted to automate that workflow?
00:30:50
Speaker
Again, I'll preface that IT t teams would hate this. This is like shadow IT, shadow AI, and none of the records that I'm uploading here are like actual site records. It's easy to ask for forgiveness, the commission, just remember that. Yeah, yeah. um So my workflow basically involves, so old I'll walk through the tools that I use for anyone that wants to know.
00:31:16
Speaker
um But effectively, I have used make dot.com, which is kind of like an automation tool and stuck together a bunch of actions that string together into a workflow. And so roughly what happens is I have it watching the Google Drive folder.
00:31:34
Speaker
um to identify when anything gets added or removed from that Google Drive folder, that could easily be a SharePoint folder as well. It will then, in the next step, download that file um and send that file in the third step to Anthropic Claude. ah Claude will, um in that step,
00:31:56
Speaker
read the read the record and identify any potential early warnings or issues in the record that we should know about. um It then sends that to an open AI API,
00:32:14
Speaker
which is going to, because I struggled to get this step to work, that's going to basically return- ICT is going to love this. let's Let's get this to every single AI platform out there all in one workflow. That's great. Well, you can then, you can't go wrong if you spread the risk. um So then I got it to basically, I couldn't get this to work easily. I got that step is basically returning a yes or a no as to whether we need to send an early warning notification.
00:32:40
Speaker
um So that's what that step's doing. Then I've got a router which basically says, does it need an early warning, yes or no? And if no, it's gonna go to Gmail and send me an email that says, I've checked it, this record's all good. And if it does need an early warning, it gets routed back to Anthropic Claude and Claude is gonna draft the early warning notice in a similar way to what we looked at in um and chat GPT.
00:33:08
Speaker
So basically every record that gets added to a folder, so you could imagine this is kind of watching the file where all the diaries go, and then it's got a pipeline to identify and notify any diaries that need to get added. So let's go on, add a file. Oli will have to black out any of this, because I think I'm sharing my screen with all sorts of stuff on it. So um make sure I'm not sharing something I should have.
00:33:35
Speaker
ah So I'm basically adding another thing to this file called site diaries. Then I head back to my workflow. I can set this to run automatically, but I'm just gonna manually run it. So you can see here it's identified that it's downloaded the file and now it's currently set with an anthropic while it's ah going through and um actioning the like, do I need a early warning off the back of this?
00:34:03
Speaker
And then it's going to pass it over to OpenAI, which is determined, yes, this does, which then routes it back to Claude, which is now writing my early warning. And then once it's done that, it'll let Gmail know, and that has pinged me an email. And so if I head back to my email, I just got an email from myself ah with an early warning draft in there and a link to the actual diary record that it came from.
00:34:32
Speaker
I built that in probably an hour. That's really impressive. I'm shook. And so yeah, once you, there's probably like a Sunday afternoon with a beer that would get you the first version of that. But once you've built it and set it
Construction Happiness Survey
00:34:46
Speaker
all up, you could, there's all sorts of workflows you can tie up the back of that.
00:34:50
Speaker
um you That cost me $1.50 in um API calls on Anthropic, and that's because it took me 25 goes to get one of the steps to work where I was sending ah a lot of documents. i'm i'm sure I'm not sure you've recently looked at the hourly rate of a QS, but it's good value.
00:35:10
Speaker
Yeah, yeah, certainly. I can't pay for that myself. A friend of a friend who is a copywriter and instead of doing any copywriting, he outsourced and paid personally for outsourced copywriters to DAWS copywriting. So his job was done by people that he would pay personally.
00:35:25
Speaker
is the equivalent of that, right? I'd definitely pay for this myself. Yeah, 100%. All right, we've given ourselves one and a half minutes to cover topic number three, which I'll briefly ah talk about, and then we'll, you'll give you a two minute thought. In summary, in late November last year, the first construction happiness barometer survey was taken, initially polling 300 construction professionals about happiness levels in the industry.
00:35:52
Speaker
and what the causes, et cetera, were. The survey revealed that 50% saw their colleagues as the biggest source of happiness, interesting, followed by the satisfaing of delivering satisfaction of delivering a project at about 50%. Salary and benefits most notably comes in at about third at 35%. Any thoughts from what you read there, Carlos, that gave you any pause for thought or reflection?
00:36:16
Speaker
The sample of like three people was the most concerning part of this report. But ah no, it's a it's definitely true. Like the best project I ever worked on by far was also the hardest work, the longest hours, but it was a good team so everyone enjoyed it. I guess if you're a project leader, taking that information of understanding that happiness is a bigger driver than cost, it should hopefully mean that you do things to keep the team engaged, do things that improve the 60 hours a week that they they are sat in that office for, um rather than just dishing out pay rises and hoping people stick around because turnover of staff is awful in construction. So that's that's my feedback. Yeah, it's like the classic saying, right? People leave ah bad managers, not bad jobs or something. Yeah, I don't think there's anything. ah there's I think there's any new use. So yeah, I guess, mate, jumping back to topic number two, i would if I was on a project today, I would take away that you can stay certainly sit there and wait for some benefit of AI to be rolled out to you from head office or from the IT team.
00:37:23
Speaker
um but you might be better to just kind of get your hands dirty and play around. There's this idea that someone says, AI, you're immediately in this big buzzword thing where you're going to be pitched something super complicated. yeah Just try the real simple stuff that helps you each day. um yeah yeah Start with a start with a time start with a like a time analysis and think about where do i spend where do I spend most my time? Where am I losing a bunch of time doing repetitive stuff?
00:37:49
Speaker
The more you play with it, the
Conclusion and Listener Acknowledgments
00:37:51
Speaker
more you'll uncover other things it'll do ah with you and for you. And yeah, I don't think you need a big SaaS application to sell you how to do AI on your project. You can start diving in straight away. absolutely Awesome. Mate, thanks for the prep work and thanks for showing some interesting things. Do you want to read us out?
00:38:09
Speaker
Thank you very much everyone for tuning into today's show. If you did enjoy the episode, please do think about liking the video or following us on your chosen podcast platform. We really appreciate your support and we'll catch you all next week. Thanks. Bye bye.