---
url: "https://youtube.com/watch?v=cmATJGbA8bI&is=Zak3y8bnO2H2-qR3"
title: I built an AI code review bot in 30 minutes - here’s how
source_kind: youtube
author: How I AI
captured: "2026-08-17T00:09:08+00:00"
comment_tree: false
topics: [ai-tools]
summary: A guide to building an AI code review bot that scores and auto-approves low-risk pull requests to reduce human review workload.
status: ok
---

# I built an AI code review bot in 30 minutes - here’s how

Channel: How I AI

## Transcript

[0s] Everybody is drowning in AI created PRs.
[3s] When anyone can code, anyone can cook,
[6s] and anyone can vibe anything to GitHub,
[10s] that means that a lot of us are sitting
[12s] on a big queue of pull requests that
[15s] then we have to review. If you are an
[18s] engineering leader or a software
[20s] engineer in a company that has embraced
[22s] AI in any way, shape, or form, you feel
[25s] this pain. And I get this question all
[27s] the time. We figured out how to code
[29s] with AI. Now, what do we do with all
[32s] these PRs? Well, people smarter than us
[34s] have figured this out. And you know what
[36s] the answer is? You don't have to review
[39s] all your PRs. Yep, you heard it here
[41s] first. You do not actually have to
[44s] review one-on-one by a human all those
[48s] PRs. So, in today's episode, I'm going
[50s] to show you how to build your own PR
[52s] review, risk-scoring, and automatic
[55s] approval bot so you can take all the
[57s] low-risk PRs off your plate, get those
[60s] to production, and focus on the ones
[62s] that really require your human
[64s] expertise. This episode is inspired by a
[67s] couple folks, especially our past How I
[69s] AI guest Brian from Intercom, who has
[73s] led the way in auto-scoring and
[75s] auto-approval PRs. But I'm going to show
[77s] you exactly how it gets done, and we're
[79s] going to get it done very quickly in
[81s] this 30-minute How I AI mini episode.
[84s] Let's get to it. This episode is brought
[86s] to you by [music] WorkOS. AI has already
[89s] changed how we work. Tools are helping
[91s] teams write better code, analyze
[93s] customer data, and even handle support
[95s] tickets automatically. [music]
[97s] But there's a catch. These tools only
[99s] work well when they have deep [music]
[101s] access to company systems. Your co-pilot
[103s] needs to see your entire code base. Your
[106s] chatbot needs to search across internal
[108s] docs. And for enterprise buyers, that
[110s] raises serious [music]
[111s] security concerns.
[113s] That's why these apps face intense IT
[115s] scrutiny from day one. To pass, they
[118s] need secure authentication, access
[119s] [music] controls, audit logs, the whole
[122s] suite of enterprise features.
[124s] Building all that from scratch, [music]
[126s] it's a massive lift. That's where WorkOS
[128s] comes in.
[130s] WorkOS gets you drop-in [music] APIs for
[132s] enterprise features, so your app can
[134s] become enterprise ready and scale
[136s] [music] upmarket faster. Think of it
[138s] like Stripe for enterprise features.
[140s] OpenAI, Perplexity, [music] and Cursor
[143s] are already using WorkOS to move faster
[145s] and meet demands.
[147s] Join them and hundreds of other industry
[150s] leaders at workos.com.
[153s] Start building today.
[155s] Okay, I know what you all are thinking.
[157s] You're thinking, "Claire, you are crazy.
[159s] Of course I have to review every PR that
[162s] comes out of AI. That's just how it's
[165s] done. We review all of our human PRs, so
[168s] I have to review all of our AI PRs." And
[171s] I am telling you, in 2026, the answer is
[175s] you can safely not review PRs by putting
[179s] AI in the loop. And I want to call
[181s] attention to two blog posts that have
[183s] really inspired me in this process. One
[185s] is from Intercom
[187s] about how they made AI-approved PRs
[190s] safe. If you don't know, Intercom has 2x
[192s] 3x their PR throughput. They are
[194s] shipping a lot of code. That means they
[197s] had a lot of code to review. And so,
[201s] they built a PR review agent that scores
[205s] their PRs and then auto-approves them.
[208s] And what I love about what Intercom has
[210s] done here is they don't just worry about
[212s] speed and velocity and volume. Yes,
[215s] they've multiplied the PRs across their
[217s] R&D organization. Yes, they want to get
[219s] those PR cycle time down and those
[222s] reviews out quickly. But what they
[223s] really think and what they've proven
[225s] with their adoption of AI is that
[227s] AI-approved PRs and AI-written code in
[231s] general can be even safer and even
[234s] higher quality than what you're doing
[236s] with just a human in the loop. And so,
[239s] they have used AI as an advantage when
[242s] it comes to safety and quality as
[244s] opposed to a disadvantage for safety and
[248s] quality. And to set the vision of what
[250s] you can expect here, I love that they
[252s] shared some metrics. One is that AI
[256s] approved PRs are approved faster,
[259s] actually five times faster than their
[261s] human PRs. So, one, you are getting some
[264s] speed. Two, they are seeing higher
[267s] quality, so the revert rate, the amount
[270s] of code that has to be fixed after it
[272s] hits production,
[274s] is much lower when they write that code
[277s] with AI. And then they figured out how
[280s] to label and trace and audit all those
[284s] changes and approvals so that they are
[286s] in compliance with all of their required
[290s] compliance frameworks, whether those are
[291s] SOC 2, HIPAA, etc. So, I often get this
[294s] objection from folks, they say, "Well,
[295s] I'm in a SOC 2 environment, I couldn't
[296s] possibly
[298s] auto approve PRs." But in fact, there
[300s] are frameworks, as long as it's in your
[302s] risk policies, as long as it's in your
[304s] code review policies, as long as it's
[305s] auditable, queryable, and defensible,
[308s] you can do this within your compliance
[310s] framework. I'll have to work with your
[311s] security and compliance teams to figure
[312s] out how this can work for you.
[314s] So, I really love just this whole
[316s] concept. I also took some inspiration,
[318s] who took inspiration from Intercom on
[320s] this, Rewind Bot, which is a little bit
[323s] more of the technical way that they
[327s] review PRs. And so, this blog post about
[331s] Diff Vader, which I love, shows a couple
[334s] components of how they've scored risk,
[336s] whether it's blast radius, how correct
[338s] it is, if all the actions have run, etc.
[341s] And so, between these two blog posts, I
[343s] had a really good starting point for
[347s] building an agent. I knew that I wanted
[349s] to do this for chat PRT. I knew we had a
[351s] lot of low-risk PRs that were just
[353s] sitting in the queue because my
[355s] colleague and I were not getting to
[356s] review them. It was kind of boring.
[359s] Especially a lot of like Devon authored
[360s] PRs. And so this was going to be a
[362s] really high-impact thing that I knew I
[363s] could build and also let me work with a
[366s] new framework that I'm excited to demo
[368s] for you all here on How I AI. So,
[373s] we're going to go to this Code X vibe
[375s] coded deck and we're going to walk
[377s] through how I built a code review risk
[380s] scoring bot with Vercel's Eve. So, if
[384s] you have not heard about Eve, I'm a
[386s] little bit obsessed with Eve. Why do I
[389s] like Eve? Well, Eve has become the
[392s] simplest way for me to deploy
[396s] AI agents in the enterprise in Slack and
[399s] GitHub
[401s] without the torture and pain of
[403s] something like Hermes agent or
[407s] my beloved OpenClaw. So, when I'm
[410s] working
[411s] and helping other companies sort of like
[413s] think about how to deploy agents, if
[415s] they're using Vercel, I tell them to use
[417s] Eve. Why do I tell them to use Eve?
[419s] Well, what's awesome about Eve is a lot
[421s] of it is just a directory. An Eve agent
[424s] is very similar to OpenClaw, just a
[427s] directory of instructions, skills, code,
[430s] and it can talk to you in channels that
[434s] are out of the box in Vercel. So again,
[437s] it's like very easy to set up the agent.
[440s] It's very easy to give it skills, which
[442s] we all use. It's very easy to give it
[444s] tools. It can even have a sandbox in
[447s] which it can execute things including
[449s] your code if you're doing a code review.
[451s] And then my favorite part is
[454s] managing talking to the agent in
[456s] specific channels and connecting it to
[459s] enterprise data is super easy. So,
[462s] because Vercel
[464s] has shipped a Vercel connectors, which
[466s] are like managed connections inside your
[468s] Vercel account, you can just literally
[470s] go through like a little wizard to hook
[472s] up Slack to your Eve agent. Little
[474s] wizard to hook up GitHub to go through
[476s] your agent. And it's all set up. The
[479s] refresh tokens, all that stuff.
[480s] Everything that's painful about creating
[482s] an agent is just taken care of. And I
[484s] sound totally like
[486s] a Vercel shill. Um I've just tried a
[488s] bunch of stuff, and this has been the
[491s] easiest framework to get up and running.
[494s] It uses the chat SDK behind the scenes.
[497s] And if you are creating any sort of
[499s] chatbot in Slack, Teams, WhatsApp,
[502s] whatever,
[504s] I mean,
[505s] bless. I have I was asking for this chat
[507s] SDK for a long time. It got built. This
[510s] is an open-source framework that I think
[512s] is so useful, whether or not you use Eve
[514s] or not, to stand up a chatbot. It just
[518s] like
[520s] takes away all the complexities of
[522s] managing like multi-channel agents. It
[525s] handles all like the Slack
[526s] configuration. It gives you a manifest
[529s] for Slack. Like I love it that Chat
[530s] Perry's Slack bot and Teams bot actually
[532s] runs on the chat SDK, but it's even
[534s] better when it's bundled into this Eve
[536s] framework. And so,
[538s] I highly recommend you try Eve
[541s] framework. Again, like they didn't pay
[543s] me to say this. I just really like it,
[545s] and I found it super easy to use. And
[546s] now every agent that I'm creating that I
[548s] want to interact with
[550s] in Slack for an internal use case, if I
[552s] don't already have an open claw, I'm
[554s] porting over to Eve. So, how did I
[558s] actually build this? What does the agent
[560s] do? Well, I built it in Codex. And so,
[564s] I'm going to show you the amazing prompt
[566s] where I started this PR review bot,
[568s] which it says, "I want to make an
[570s] internal GitHub bot/app that reviews PRs
[572s] after all CGXs, that's checks, are green
[576s] and grades low, medium, high risk and
[578s] approves the low risk Oh my god, you
[580s] guys I can't type. All All of the type
[582s] of this. Approves the low risk PRs
[584s] automatically. That's all I did and then
[588s] I interrupted it. I steered it and said,
[589s] we could design it as a Vercel Eve agent
[592s] if you want and it said, sure that seems
[595s] like a good good idea and it was off to
[597s] the races.
[598s] And so instead of going through kind of
[600s] like the chat on Codex on how I built
[603s] this is took a couple turns but really
[604s] not that many. I would say a couple
[606s] things that were like magical as I ran
[610s] Codex is not that it did the code for
[613s] the Eve agent in the repo. That was
[614s] fine. That was interesting and I'll tell
[616s] you a little bit about how the code
[617s] actually works in a sec. But what was
[619s] actually awesome is if you've ever set
[621s] up a Slack bot or a GitHub app, you have
[625s] to like click through a bunch of
[627s] configuration screens. You have to pick
[628s] permissions. You have to do all this
[630s] stuff. And so what I did is my favorite
[634s] hack is I had it use Chrome browser use
[637s] and I let it navigate Slack bot setup. I
[640s] let it navigate
[642s] um GitHub setup and I just like clicked
[645s] buttons and two-factor and just
[647s] double-checked that it was right. And so
[650s] I said, you know, is it configured yet?
[651s] It said, no, it needs to configure all
[653s] this stuff in the permissions on the
[654s] GitHub bot. I was like, great, you're
[656s] logged into Chrome and then I was like,
[659s] I can press save. It was having a hard
[661s] time press save. And so if you're just
[663s] sort of like meta tip, if you ever have
[665s] to do like complex SaaS configuration,
[668s] set up an app, you're like, uh, I'm fine
[671s] writing code but I don't want to do the
[672s] configuration in somebody's third-party
[674s] service, using Chrome browser use is
[677s] such a useful hack to get that stuff
[680s] done. So again, I just built this in
[683s] Codex. I actually didn't give it any
[684s] instructions on scoring. I didn't give
[686s] it any instructions on configuration.
[690s] I didn't give it any instructions on
[692s] risk. I just said kind of like one-ish
[694s] shot it, few shot this into production.
[697s] And so, I want to talk a little bit
[698s] about what the agent does and how it
[700s] works. So,
[702s] it's pretty simple. If you were building
[705s] PR review agent, this is what I would
[707s] success suggest you do, which is it
[710s] reads the PR, it looks at the exact
[713s] diff,
[714s] um and it scores the risk and it
[716s] publishes the evidence to the risk. If
[718s] it's low risk, if it's assessed that
[720s] it's a low risk PR, it submits an
[723s] approval on the PR. If it needs a human,
[727s] it escalates to human and if there's a
[730s] blocking um requirement,
[733s] then it stops and it it will require a a
[737s] request change. So, it's like very
[738s] similar to how a human would do code
[741s] review. How does it work from a
[743s] technical flow perspective? So, the
[745s] GitHub I have a GitHub app. It keys off
[749s] an event after all of the PR changes are
[752s] done. Vercel's GitHub integration picks
[755s] that up in the GitHub Vercel channel and
[758s] it sends it a little bit of information.
[760s] Vercel spins up its sandbox, it checks
[763s] out the repo, it runs it and looks at
[765s] the diff, it runs a couple skills and
[768s] tools to assess the risk and quality of
[771s] the PR, and then on the end it outputs
[774s] like a thumbs up check uh like it needs
[776s] changes and a comment. one fun thing
[779s] that it does is after it does that
[780s] review, it pings my colleague and I in
[783s] Slack to say this PR is ready for review
[785s] or it needs some help.
[788s] So, if you were to set up your Eve agent
[790s] to be similar to mine, what would it
[793s] have? Well, it would have a couple
[795s] things. It needs to have the GitHub um
[799s] channel, it needs to have instructions,
[801s] which I can show you a little bit of the
[802s] instructions. It needs a skill to review
[805s] the PR, it needs two tools. Read like
[809s] all the files and information around the
[811s] PR and then how to actually put a risk
[814s] decision in and then it has, you know, a
[817s] couple things like a Slack notifier. So,
[819s] this is not that many files. It's pretty
[821s] simple. I will show you our instructions
[824s] just to make this like very clear how
[826s] easy it is to set this up. It's a chat
[828s] parity engineering agent. It reviews
[831s] PRs. It calls the risk context. It
[833s] scores it and then it has a couple of
[836s] instructions. This is it. There's no
[837s] scrolling. It is like four four or five
[840s] paragraphs, a couple bullet points, and
[842s] it's ready to go. You do not need to
[844s] over-engineer this. And it works really,
[847s] really well. So, this is again why I
[848s] like this Eve agent framework because
[850s] you're just working in natural text or
[852s] in TypeScript, and it's super easy to
[855s] implement.
[857s] Setting up the GitHub app, you can
[859s] ignore this. I mean, I guess like point
[860s] your agents to this instructions. I did
[863s] not do any of this. Again, Chrome
[865s] browser use did this on our behalf. And
[867s] so, it just has to have access to the
[870s] pull request, the contents of your
[871s] files,
[873s] um CI checks and action checks, and then
[876s] some metadata. And I just had Vercel set
[879s] all that up. It has a couple rules on
[883s] when it's triggered. So, you don't want
[884s] it triggered on every PR before checks
[886s] are done. So, I kind of had to like
[887s] reduce some noise in the PR rules. And
[891s] then there's just two basically steps to
[893s] this.
[894s] It reads the PR.
[897s] It's like a skill to read the PR. I will
[898s] show it to you. It's not that thrilling,
[901s] but again, just showing you how simple
[904s] it is.
[905s] It reads and reviews the PR. It has a
[910s] couple repo specific
[912s] uh risk kind of category. So, like docs
[915s] are low risk, feature log it logic is
[918s] medium risk, and then like
[919s] authentication billing's high risk. Diff
[922s] size does not set risk, and then it
[925s] gives some like instructions on writing.
[927s] Again, this is it. It's like not that
[929s] much. It's maybe one page of text. I did
[931s] not write any of this. I refined it. And
[933s] so, what I'm trying to reinforce to you
[935s] is like writing these agents is not that
[937s] hard. It is literally just writing
[939s] instructions and skills. That is what it
[941s] takes to create an agent like this, and
[943s] that's why I like this e framework cuz
[944s] you can do a lot of it in markdown.
[947s] And then once it reads the code, it
[950s] scores the risk. And so, it looks at six
[952s] things: how big is the change surface
[955s] and blast radius, is it easily
[957s] reversible, right? Like a huge data
[959s] migration is maybe harder to reverse.
[962s] Does it touch data security and does it
[963s] cover data security? Does it change
[965s] anything about our operations? And
[967s] verification gap is like, are our tests
[969s] complete? Did CI complete? Can we
[972s] actually test a couple ways that this is
[974s] true?
[975s] And then it has sort of like a script
[978s] that it runs to calculate a score. And
[980s] then anything under 24 points, I did
[982s] again, I did not like choose these
[983s] thresholds. Anything below 24 points is
[985s] low risk, medium risk is 25 to 64, and
[988s] then 65 and above is
[990s] high risk. Medium and high risk PRs
[992s] require human
[994s] approval. Um so, it's pretty simple.
[998s] There's some logic in here to only
[999s] review the latest changes, which I like.
[1003s] And then the final piece is it escalates
[1006s] us to us in Slack for final action. So,
[1008s] we have repo rules probably very similar
[1010s] to all of you that requires
[1013s] a review in our GitHub to like match all
[1016s] of our our SOC 2 compliance
[1018s] requirements. It just makes auditing and
[1021s] um management of our compliance really
[1023s] easy.
[1024s] I went through a couple ways where the
[1027s] bot could like {quote} act like a human
[1030s] to do the review. It just didn't seem
[1031s] worth it. So, what it does is escalate
[1034s] to Slack, and then it says this one's
[1035s] been automatically reviewed. You can
[1036s] punch the approve button and merge and
[1039s] you're good good to go. And so we sort
[1041s] of like built in an operational process
[1043s] here as well. But, let's see what this
[1047s] actually looks like in action. So, I
[1050s] have three PRs for you
[1053s] um just to show you kind of how this
[1055s] works. Now,
[1056s] the the PR bot called Merge Mommy, she's
[1059s] Merge Mommy the PR review bot. Um you
[1061s] know, when you make internal bots, we
[1063s] don't have Diff Vader, we have Merge
[1065s] Mommy cuz we're fun at Chat PRD.
[1067s] And this was a docs-only update
[1070s] automatically done by Devin. And
[1072s] typically, I just like smash the
[1074s] approval button on these, merge them,
[1076s] but I found that like we were just
[1077s] getting queued up of a lot of docs
[1079s] changes that didn't really need a bunch
[1081s] of review. And so, what happens on um
[1085s] once the PR is ready to go, Merge Mommy
[1088s] runs, it gives a risk score
[1091s] six out of 10, it's very low risk
[1094s] because it only changes docs, but it's
[1097s] not auto-approved because it has merge
[1100s] conflicts. So, that's one of the scoring
[1102s] issues it has to check. And then there
[1105s] are some detailed notes here about why
[1109s] it's blocked and the approval blocker is
[1112s] the pull request has merge conflicts. A
[1114s] cleaner example here is the stock which
[1117s] I already merged, which again went
[1119s] through and said it's seven out of 10
[1121s] low risk, it's auto-approved, and you
[1123s] can see up here Merge Mommy gave a
[1125s] little check that said it's approved.
[1127s] Now, what we do with these approved PRs,
[1129s] as I said, is we send them to a Slack
[1132s] channel. And so, the Slack channel says,
[1134s] "Hey, either of you can take this,
[1137s] approve it. The risk is low, the checks
[1139s] are green. You just need to smash that
[1141s] approve button and merge it." So, it's
[1144s] just like a really nice automated flow.
[1147s] Now, why do we have to have a human do
[1149s] it? As you can see, this check mark is
[1151s] gray. A bot cannot actually meet our
[1154s] repository rules for required approval,
[1158s] and so we can either skip those rules,
[1160s] hack around it. We just decided this
[1163s] check would be a signal that our humans
[1164s] can approve it without really looking at
[1165s] it and then merge it as we want.
[1169s] And then let's show one where Merge
[1171s] Mommy did not approve it. This is a
[1174s] deprecation PR, so we had chat V1 and
[1177s] chat PRD and we're moving to chat V2. We
[1179s] kind of like refactored a bunch of code,
[1181s] but we left the old stuff behind a
[1182s] feature flag, and it was like ready to
[1185s] delete. So there's like 35 changes, a
[1187s] big red diff that we wanted to delete,
[1190s] and you can see here
[1193s] Merge Mommy like I'm sorry, I can't take
[1195s] myself seriously when I say it, but I'm
[1197s] going to say it anyways. Merge Mommy
[1199s] gave it a 45 out of 100, a medium risk,
[1201s] and it was not approved by a human
[1205s] because both it had some code issues as
[1209s] well as the policy said it changed
[1212s] server API behavior. It was a large
[1214s] change service, and so it's medium, not
[1217s] low, and it can't approve the PR. Now,
[1221s] we can run this against every PR that
[1224s] goes through our process, and again,
[1226s] we're going to just keep getting these
[1228s] notifications in Slack
[1231s] that just let us know, "Hey, this can be
[1233s] approved or this requires human review."
[1236s] Again, I like to say this thing where we
[1237s] can put AI to work for us or we can have
[1241s] AI put us to work, and this is an
[1243s] example where you do both. I put AI to
[1246s] work, this lovely Eve agent, I put AI to
[1250s] work to review and score
[1254s] the PR and tell me if there's any risk
[1256s] and like be detailed eyes on it, and
[1259s] then I have AI put me to work by
[1261s] escalating into Slack for the final
[1263s] action and making it like a two-click
[1266s] process to get this thing to production.
[1270s] Now, I just have to say this is
[1272s] something that I was really intimidated
[1275s] to build. I thought this was going to
[1277s] take me like days and days and days. I
[1280s] was like, "I don't want to set up the
[1281s] GitHub app." I actually had tried this
[1284s] before Codex like browser use was really
[1286s] good and before the Eve agents had come
[1289s] out and it was just like not easy to do.
[1292s] But now that we have these open-source
[1293s] frameworks, now that we have these
[1296s] agent-building frameworks, now that kind
[1299s] of like everybody's agreed that we need
[1301s] Slack bots that do AI,
[1303s] a lot of these problems have been solved
[1305s] and then Codex browser use was just like
[1307s] the cherry on top that made building
[1310s] this agent super super super simple. So,
[1314s] to just recap for you all in this
[1317s] lightning-fast mini episode about how to
[1320s] build a PR risk score and review bot,
[1324s] use Vercel's Eve agents, give it
[1327s] instructions and a skill to read a PR
[1330s] and calculate a score based on a couple
[1333s] components,
[1334s] hook up that Vercel agent to GitHub and
[1337s] to Slack, let it read from GitHub and
[1340s] give you a little gray check for
[1342s] approval and then have it Slack you in
[1345s] Slack for that final human review and
[1348s] merge, and your cycle time will be super
[1352s] super fast and your PR velocity will go
[1355s] through the roof. Now,
[1358s] I want to hear from you all. Is this
[1360s] crazy? Can you pull this off at your
[1362s] company and what would you put into your
[1365s] risk scoring that I haven't thought of?
[1369s] The final bonus piece off this, which I
[1371s] didn't show, but a lot of great folks
[1373s] do, especially the the folks at Intercom
[1376s] do as part of this process, is they run
[1378s] evals on this internal agent. So, every
[1382s] time this review is run, it gets logged
[1385s] into I think it's like an internal eval
[1386s] platform, and an engineer looks at it
[1389s] and says, "Did the agent get this right?
[1390s] Did the agent get this wrong? Are we
[1392s] happy with the scoring mechanism?" So,
[1394s] very similar to how you'd use evals to
[1396s] improve your customer-facing AI
[1398s] products, you're going to want to use
[1400s] evals to improve your internally-facing
[1403s] AI bots, especially ones that touch
[1405s] really critical things like code. That
[1409s] has been my mini episode on how to build
[1411s] a PR approval and review bot here at How
[1414s] AI. I cannot wait to hear if you think
[1417s] this is crazy or this is the way of the
[1420s] future. Thank you so much for joining,
[1422s] and we'll see you soon.
[1425s] Thanks so much for watching. If you
[1427s] enjoyed the show, please like and
[1429s] subscribe here on YouTube, or even
[1431s] better, leave us a comment with your
[1432s] thoughts.
[1433s] You can also find this podcast on Apple
[1435s] Podcasts, Spotify, or your favorite
[1438s] podcast app. Please consider leaving us
[1440s] a rating and review, which will help
[1442s] others find the show. You can see all
[1445s] our episodes and learn more about the
[1447s] show at howiaipod.com.
[1450s] See you next time.