Skip to main content
Shyft

I'm Daan, building Shyft from Costa Rica and Portugal

I run Shyft with a team of four AI agents. I approve, they do the work.

Helix keeps the plan, Arc runs engineering, Chase ranks what moves revenue, and Scout watches the stack. Nothing that costs money, hires someone, ships a high risk change, or reaches a customer happens without my approval.

Request an invite →Meet the team →

709

pull requests merged across my repositories

last 30 days · measured 14 Oct 2026

7

repositories with at least one merge

last 30 days · measured 14 Oct 2026

A script in the repo counts these on GitHub. Every pull request is opened under one account, so this does not say who wrote the code.

The team

Meet the four agents

Each one has a clear job and reports with evidence.

Helix, Chief of Staff

3 tool calls
  • read_boardgoals and milestones loaded
  • draft_digestmorning digest drafted
  • route_approvalsent to Daan on Telegram

Example run

Owns the operating rhythm. Goals, milestones, the weekly goal report, and the morning digest to me.

  • Report numbers come from measurement scripts in the repo, never typed by hand.
  • A missing number prints as an explicit "awaiting real data" marker instead of a guess.
  • Triages the stale backlog.
  • Routes approvals to me.

Arc, CTO

3 tool calls
  • read_github_issueissue and brief loaded
  • dispatch_coding_agentPR opened by a coding agent
  • request_reviewa different model family reviews the PR

Example run

Owns engineering. Picks the next work from GitHub issues, writes the brief, and dispatches it to coding agents.

  • Coding agents run in their native harnesses, Claude Code, Codex, and Gemini.
  • Product leads under the CTO supervise each product.
  • Every change gets an independent review by a different model family than the one that wrote it.
  • Reports carry the exact commit, the command run, its exit code, and verbatim output. A claim without evidence is reported as unknown.

Chase, CCO (Chief Commercial Officer)

3 tool calls
  • read_billing_and_usagebilling, usage, and shipped work loaded
  • rank_leversranked by expected revenue effect, arithmetic shown
  • wait_for_choicewaiting for Daan to choose

Example run

Answers one question. What moves revenue next, and what is the evidence.

  • Reads billing, product usage, and what shipped.
  • Ranks levers by expected revenue effect with the arithmetic shown.
  • Ranks and recommends. I choose.
  • Outbound sends run under a hard weekly cap that a script checks before every send.

Scout, Stack and AI radar

3 tool calls
  • read_changelogsstack releases and provider news read
  • file_proposalproposal filed, marked trial
  • send_to_arctechnical proposal sent to the CTO

Example run

Reads changelogs and releases for the whole stack, AI provider news, Hacker News, and GitHub trending every day.

  • Files at most three proposals a day.
  • Each proposal is marked adopt, trial, watch, or drop, with effort and risk.
  • Technical proposals go to the CTO.
  • Anything with spend, a new vendor, or a strategy change goes to me for approval.

How it runs

Paperclip holds the org chart. GitHub holds the work.

I keep the approvals.

The stack

  • Hermes Agent runtime. Each agent has its own profile, credentials, and memory.
  • Paperclip Orchestration. Holds the org chart, goals, approvals, routines, and the cost ledger.
  • Claude Code Coding harness.
  • Codex Coding harness.
  • Gemini Coding harness and model family.
  • GitHub Holds the work. Issues, pull requests, and CI.

Approvals and evidence

  • My agents wake on assignment, not on a timer.
  • I approve spend, hires, high risk merges, and anything a customer sees.
  • Approvals reach me on Telegram.
  • Low and medium risk changes merge after independent review and green CI. High risk changes wait for me.
  • A claim without evidence is reported as unknown.

The offer

I set up the same team for your business

I run Shyft with this team. I set up the same structure for you.

  • The same seats, adapted to your business.
  • Running on your own accounts and credentials.
  • Approvals go to you.
Request an invite →

FAQ

Before you ask

What does the agent team actually do?

Four agents run the operating side of Shyft. Helix keeps the plan and the morning digest, Arc runs engineering, Chase ranks what moves revenue, and Scout watches the stack. Each one reports with evidence.

What do you set up for me?

I set up the same seats, adapted to your business. They run on your own accounts and credentials, and approvals go to you.

Do I need a technical team?

No. You approve and the agents do the work. The reports are in plain English.

Whose accounts and credentials does it run on?

Yours. The agents run on your own accounts and credentials.

What do the agents never do without approval?

Spend money, hire, merge a high risk change, or do anything a customer sees. Those wait for your approval.

How do I start?

Request an invite. I read every request and follow up.

Our own products

We also build our own products

Independent products with their own accounts, credentials, and customer data. Two are still in beta.

Rotor

Beta

Durable TypeScript workflows

Retries, timeouts, and state handled across steps. Queues, crons, and human approval gates inside a run.

Request an invite →Visit Rotor →

Crispy

Live

LinkedIn automation for AI agents

Search prospects, send connection requests, post content, manage conversations, and track engagement.

Visit Crispy →

Zend

Beta

Cold email infrastructure API

Domains, inboxes, warmup, and sending behind one API. Built to run outbound at volume.

Request an invite →Visit Zend →

ranq

In build

Visibility in AI search

Tracks whether generative engines name and cite your brand on the pages that decide your category.

Vibes

Vibes, the community

Vibes is where we work in the open with other people building AI into their operations. You could be part of it.

Join Vibes →
What happens inside →