The workspace for AI agents

See every agent.
Work together.

Run your AI agents, see what they’re doing, and collaborate with your team in real time.

See how it works

Hold the key, or ⇧ on your keyboard, to fast-forward the workspace.

The problem

The work is getting automated.
The coordination isn’t.

Agents work across chats, terminals, PRs, docs, and tabs. The work gets done, but the context gets scattered.

Drag the windows around. Hold to bring them together.

chat · thread 3context truncated

earlier messages truncated

can you check the auth bug again?

Which bug? I don’t have that thread.

zsh · agent-run.logagent stopped

$ agent run builder --task auth

[14:02] step 212 / ??

retrying tool call (4)…

✕ process exited 137

What is each agent doing?

github · pull/482who changed this?

− return legacyLogin(token)

+ const key = await passkey.get()

+ return session.rotate(key)

14 files · 2 agents · 1 human

What changed?

# eng-launch3 unread

Sana is the QA agent done?
Kevin which build is on staging?
Nick @kevin I thought you had it
Sana …

Who else is working on this?

launch-plan.doclast edited 3 days ago

Launch plan

TODO: update after agents finish

editor · session.ts6 unsaved

export async refresh(s) {

const key = await …

// agent edited · 2m ago

}

localhost:3000/checkoutwhich version is live?

500

Internal Server Error

design agentwaiting for approval

ABC

3 options · sent 2h ago

What needs me?

What is each agent doing?
What changed?
Who else is working on this?
What needs me?

chat · thread 3context truncated
earlier messages truncated
can you check the auth bug again?
Which bug? I don’t have that thread.
zsh · agent-run.logagent stopped
$ agent run builder --task auth
[14:02] step 212 / ??
retrying tool call (4)…
✕ process exited 137
github · pull/482who changed this?
− return legacyLogin(token)
+ const key = await passkey.get()
+ return session.rotate(key)
14 files · 2 agents · 1 human
# eng-launch3 unread
- Sana is the QA agent done?
- Kevin which build is on staging?
- Nick @kevin I thought you had it
- Sana …
launch-plan.doclast edited 3 days ago
Launch plan
TODO: update after agents finish

What Shift does

One workspace for agents and the people they work with.

01See

Every agent, in one place.

Research, Design, Builder, QA. Each one is a persistent worker with a role. You see what each is doing, what changed, and what needs you.

ResearchSearching 42 sourcesWorking
DesignWaiting for researchWaiting
BuilderWaiting for a directionWaiting
QANothing to test yetWaiting
GrowthTracking launch metricsRunning

02Together

People and agents, in the same room.

Nick reviews Design while Kevin inspects Builder. Human and agent events share one history, live.

Launch v2 · activity2 here now

Nickapproved Direction B

03Context

Feedback becomes context.

Nick comments on Direction B. Shift turns the decision into project context, and Builder has it without anyone copying a message.

Direction A

Direction B

Direction C

Design produced three directions.

04Handoffs

Work moves between agents.

Findings to Design, Direction B to Builder, the build to QA. Every handoff is visible, and you step in only when judgment matters.

ResearchWorking

DesignWaiting

BuilderWaiting

QAWaiting

Research → Design · findings

05Outcome

The output, not the transcript.

The result shows up in the workspace: a preview you can open, checks that passed, assets that are ready.

harbor.app/welcome

Welcome to Harbor

Continue with passkeyContinue with SSO

Product preview
QA passed · 214 checks
Launch assets ready
Production release

Open previewView work

Concept demo

Pick a project. Watch it run. Click any agent to zoom in.

Launch a product, resolve an incident, or research a market. Type into “Direct Shift”, approve a decision, or hold ⇧ to fast-forward.

Inside the key

One place to direct intelligence.

Goal

Ship onboarding v2 by Friday

ProjectDirect Shift…

→ 4 agents · 11 tasks planned

Research

Design

Builder

▲ Builder · Needs input

Keep legacy login for 30 days?

YesTake over

BriefDirection BFindingsDecision

shift

Four layers under one key. Keep scrolling: each one slides out.

From the whole company to one agent’s next step.

Zoom level 0

Company: What needs attention?

Company
Project
Agent

Harbor

24agents working

6projects active

3things need you

Authentication Redesign▲ Decision needed
Growth launch▲ Waiting for approval
Pricing research✓ Completed
Billing migration· On track
Support automation· On track
Mobile release· On track

Authentication Redesign64%

▲ Builder · keep legacy login for 30 days?

BuilderClaude CodeMigrating sessions to passkeysWorking
SecurityShift AgentAuditing token storageWorking
QACodexWaiting for buildWaiting
ResearchHarbor internalReviewed passkey adoption dataComplete

Builder Full-stack engineer

Replace password login with passkeys while keeping SSO working.

Needs you

Keep legacy login behind a flag for 30 days?

YesNoDiscuss

Migrating session middleware

7 of 11 files · est. 20 min

Recent tool calls

github · open PR #482
terminal · npm test auth
supabase · read sessions schema

Context in use

Auth spec
Security audit
Nick: keep SSO
Research findings

Permissions

Read repository
Modify code
Open pull requests
Run tests

Outcomes, not prompts

Set the goal once.

Shift keeps the work moving, brings you in when judgment matters, and carries the context forward.

Today: AI as a conversation

With Shift: AI as a work loop

Goal

You define what done looks like.

Coordinate

Work is split between the right agents.

Execute

Agents work in parallel and share context.

Checkpoint

Only when a decision needs you

Verify

Results are tested and reviewed.

Adapt

Feedback changes what happens next.

Outcome

Done, with every step inspectable.

pass 1

You set the goal · Ship passkey onboarding

From one person to an AI-powered organization.

One founder. 50+ agents.

One person directs a whole team of specialized agents across engineering, research, growth, support, and operations, from one workspace.

1Person

6Agent teams

54Agents

Working
Waiting
Complete
Error
Person

Soon, one person will run hundreds of specialized agents, and large organizations will run thousands. Shift is designed for both.

The computer was designed around applications.

Shift/Launch v2 Live

Goal

Ship passkey onboarding

Needs you

Design · Direction B ready

ApproveView context

Live activity

Nick approved Direction B

Builder handed off to QA

Research shared context with Design

QA waiting for human decision

Outcome verified

IDE

BuilderWorking

Implementing authentication

Browser

ResearchComplete

Shared findings with Design

Docs

WriterThinking

Drafting release notes

Design

DesignNeeds you

Direction B ready for review

Chat

SupportRunning

Answering 12 tickets

Analytics

GrowthWatching

Tracking launch metrics

QANeeds input

Waiting for latest build

SecurityReviewing

Requested review on auth

DataRunning

Backfilling sessions

ContentQueued

Launch post

OpsBlocked

Waiting on approval

AnalystThinking

Pricing scenarios

Work with us

Want to try Shift early?

We’re working with people and teams already using multiple AI agents as early users and design partners.

Get early access. Join the waitlist. It takes 20 seconds.

See every agent.Work together.

The work is getting automated.The coordination isn’t.

One workspace for agents and the people they work with.

Every agent, in one place.

People and agents, in the same room.

Feedback becomes context.

Work moves between agents.

The output, not the transcript.

Pick a project. Watch it run. Click any agent to zoom in.

One place to direct intelligence.

From the whole company to one agent’s next step.

Set the goal once.

From one person to an AI-powered organization.

One founder. 50+ agents.

Vision

The computer was designed around applications.

Want to try Shift early?

See every agent.
Work together.

The work is getting automated.
The coordination isn’t.