Episode 3 3:47 Agents, Repository, Discussion
The Impossible Ticket The night before the MVP demo, Sam is stuck on the biggest ticket. Maya points him to Helpi Agents: they plan, he approves, they code and open a pull request that he reviews.
With Sam, Maya, Jordan, Lina, Dan, Mr. Bennett, and Rachel 0:02 Daily scrum. Yesterday, today, blockers. Jordan? 0:05 Sign-in is done. Today, the magic link. 0:07 Order history is in review. Today, invoices. 0:09 Returns are moving. Today, the return label. Easy. No blockers. 0:13 Tomorrow at ten, Larkspur sees the MVP. Let's make it count. 0:25 Don't stay too late, champ! 0:29 A label, a barcode, the tests... by ten tomorrow? Impossible. 0:37 Maya, sorry, it's late. The return label: 0:39 the PDF is blank, the barcode prints upside down, and the tests are red. The demo is at ten. It's impossible. 0:44 Impossible? Have you tried the agents? 0:47 The... agents? 0:48 Helpi's coding agents, in Octopora. Give them the ticket. 0:51 They plan, you approve, they code, and they open a pull request. 0:55 And who reviews it? 0:56 You do. Remember when you were blocked, waiting for my review? 1:00 Maya, can you review my pull request? I'm blocked. 1:02 Tonight, you're the reviewer. 1:04 Show me. 1:05 Share your screen. Now open Helpi Agent. 1:09 Meet Helpi Agents: a coding team for your tickets. 1:12 Engineer, reviewer, designer, tester and docs. 1:15 Each role changes what it's told, the tools it holds, and what it delivers. 1:20 Add as many as you need, name them 1:22 , and switch each one on or off, project by project. 1:25 Each one runs on Automatic, the models that did best on the agent tests 1:29 , or on one model you pick, or on your own AI account. 1:33 Skills teach them how your team works, from Octopora or from your repo. 1:37 An agent reads one only when it's relevant. 1:41 Rules set the limits: run length, runs a month, runs at a time, tokens. 1:44 And the checks its work must pass. 1:47 Start a run, and it clones your repo into its own sandbox in the cloud. 1:51 It reaches only your code host and package registries, and it's removed when the run ends. 1:57 It reads the code, then writes a plan: goal, approach, changes, risks, and how to verify. 2:03 Nothing is written until you approve it. 2:05 Then it codes, and runs your checks. 2:08 If one fails, it fixes its work and runs them again. 2:11 Every step lands in its logs as it works: 2:14 each model call, each file it reads, each command, and the tokens it spends. 2:19 Then it pushes a branch and opens a draft pull request. 2:22 Your reviewer agent reads it and posts its findings. 2:25 One click, and the fix lands on the same pull request. 2:29 And it never merges its own work. You do. 2:44 Sam? Still here? What are we watching? 2:46 The agents. 2:47 Secret agents? 2:49 Coding agents. 2:52 Who's winning? 2:54 We are. 3:12 Good morning, Rachel. Welcome to the MVP demo. 3:14 Morning, everyone. I can't wait to see it. 3:17 Sam, can you share your screen and show us the demo? 3:19 Sure. Here's your new customer portal. 3:24 Let's start a return... and print the label. 3:28 That's exactly what we needed. 3:32 The label too? Overnight? 3:34 I had a team. 3:40 Helpi Agents, in Octopora. They code. You decide. The modules in this episode Each one has a short tour of its own.