Agents can be multiplied. You can't.
Volgenic gives you a first mate. It runs your crew of coding agents and brings you only the decisions that are yours, drawn so you can take them in at a glance.
You started using agents to get more done. Somewhere along the way, you started working for them.
Every agent asks its own questions, in its own tab. Which repo? Which branch? Which database? You end up remembering which tab had the migration, and answering questions instead of doing your own work.
Same facts. The format decides what they cost you.
Here is one update, the way an agent usually sends it, and the way your first mate does: the answer first, the change shown, the checks in a table, and the one decision a click away.
✔ Done w/ ON-resend (branch on-resend-3, PR #1187). Changes: order_notification_worker.ex — retry w/ exp backoff (base 60s, max_attempts 3, jitter 0.2, ~10 min total); order_notification.ex reuses attempt_count (no migration needed after all, see my earlier note re: 0042); 6 new cases in order_notification_worker_test.exs; ORDER_NOTIFICATION_MAX_ATTEMPTS env (default 3).
mix test: 48 tests, 0 failures (2 skipped, the billing_sync flake, unrelated). CI on main was red earlier from the ESLint bump (#1179), not this branch; rebased onto 4c1e9a2, green now. Kept the HTTP client wrapper as is per your note (4 files total).
Replay: 214 failed order notifications in the last 7d, resent from staging-2 → 211 delivered, 3 still 410 Gone (endpoints removed by the merchant, expected). Dashboard screenshot at /tmp/on-resend-3/shots/notifications-after.png.
Lmk if I should merge + deploy to prod now or hold for the Friday release train; ON-resend-2 also depends on this so either way works, also happy to bump max_attempts if you want.
Somewhere in here is the one decision that needs you.
First mate
Failed order notifications now resend themselves.
When a notification to a customer's system fails, it's tried again three times over ten minutes. I also resent last week's failures.
Before
After
| Check | Result |
|---|---|
| Tests | ✓All 48 pass |
| Size of the change | ✓Small: 4 files, no database changes |
| Last week's failures | ✓211 of 214 resent and delivered |
Ship it now?
| An ordinary chatbot | Volgenic |
|---|---|
| A wall of text, with the decision somewhere inside it. | The answer first, then the change shown and the checks in a table. |
| You remember which tab had it. | Something that already knows where everything is. |
| You scroll to find the question. | The decision is one click. |
Pictures, tables and buttons, right in the reply.
A terminal shows text, one line after another. Your first mate draws what it found in the reply itself, not linked from somewhere else, so the layout tells you where to look.
-
A player
Watch the change, or listen to it, without leaving the chat.
-
A table, and a chart drawn from it
Numbers you can scan, not numbers you have to remember.
-
Something to try, not just read
A small working app in the reply. This one works: pick a colour.
-
Options side by side
The recommended one is marked, never picked for you.
-
A review set
Keep or reject each candidate, then lock in your picks.
Everything it shows you stays one click away.
Your crew, what needs you and everything shown to you sit in a panel beside the chat. Open a linked page or work item and it opens in a pane beside the conversation, so you never lose your place.
One inbox for everything waiting on you.
When several things need you, you don't dig through chat sessions to find them. Needs attention gathers them across all your work, each one quick to understand and one click to answer. Answer them when you're ready, and go back to your own work.
Only your decisions reach you.
The questions still come. They just go to the first mate. It settles what it can within the limits you've set, and logs every call it made for you, so you can skim them afterwards. What reaches you is a choice that's yours to make.
Reach it anywhere.
Talk to your first mate on the web, or from Telegram, Slack and Discord. It also merges, builds and deploys.
- The web
- Telegram
- Slack
- Discord
Your AI agents aren't the bottleneck. You are.
Seven minutes from the founder on why one person, not the number of agents, is the limit, and how a first mate changes what reaches you.
Agents can be multiplied. You can't.
The measure is how many agents one person can realistically manage. Volgenic's job is to raise that number.