Phoenix Horizon, Inc.  ·  Palo Alto, California

Services are becoming software.We build what they run on.

The $4.6 trillion services economy is being rebuilt as software — AI-native firms that sell finished work, delivered by fleets of agents. Phoenix is the infrastructure underneath, built on two commitments: it stays neutral — any model, any harness, any cloud or the client's own boundary, no lock-in — and it makes every agent measurably better each run, proven by evals, not just watched on a dashboard.

Vendor-neutral by defaultAgents that improve every runRetains nothing by default
SCROLL
The shift

Software ate the tools budget. Now it comes for the work itself — the $4.6 trillion the world spends on services every year.

The research, the selling, the writing, the building — the work people do with their tools — has always dwarfed the tools budget. Agents collapse that boundary: firms stop selling hours and start selling outcomes. A new kind of firm is being built on that shift, and every one of them needs the same machine underneath.

SaaS
Services as software
Clients buy
a tool, per seat
the finished work
Who does it
their team
agent fleets
They review
dashboards
the deliverable
Priced by
licenses
outcomes
Where we're aimed — share of global services spend
Software & IT deliveryFirst wave
$1.5T
Research, analytics & diligenceFirst wave
$0.9T
Sales & go-to-marketFirst wave
$0.8T
Finance & accounting ops
$0.5T
Legal & compliance
$0.4T
Creative & content
$0.3T
Other professional services
$0.2T
Illustrative split of the ~$4.6T global business-services spend. Gold, lime, and emerald mark the segments AI-native firms are rebuilding first — the firms Phoenix builds for.
Who we build for

AI-native firms that sell the finished work.

01 — Research & diligence firms

Markets mapped, accounts scored, dossiers written by agent fleets — with the depth no analyst has time for.

On Phoenix: fleets researching, clients signing off.

02 — Pipeline & GTM firms

Buyers found, researched, and reached with outreach drafted from real evidence — reviewed before a word goes out.

On Phoenix: qualified pipeline as the deliverable.

03 — Software delivery firms

Agent fleets that plan, build, review, and merge — features landed in client repos with tests, docs, and a review trail.

On Phoenix: merged PRs, running in production.

The Stack

Everything under an agent-run firm.

Two years of building, end to end — so a firm selling finished work never has to build its own infrastructure.

03
Client review & approval

Rush

Where your clients and operators watch work in progress, request changes, and approve deliverables before they ship.

02
Agent orchestration

AGI CLI

One operator runs hundreds of agents in parallel — any model, any harness, sessions, credentials, and versions managed.

01
Neutral runtime · evals · audit

Prix

A neutral runtime on any cloud or the client's VPC, hosted memory, and evals that run in place — every agent proven, improved, and audited, retaining nothing by default.

Rush · Client approval
AGI CLI · Orchestration
Prix · Neutral runtime + evals
Client review & approval

Rush

See everything. Sign off on anything.

Your firm's calm surface — operators watch agents work, approve what goes out, and clients pick up finished deliverables. Real output, not another chat window.

PlatformmacOS · nativeControlapprove / revoke, alwaysShips onPrix + AGI CLI
Open Rush →
Rush — Today
Workspace
Today5
Agents12
Approvals1
Memory
Automations
Wednesday4 running · 1 waiting
Reconcile Q3 vendor invoicesRUNNING
step 7/11 · matched 41 line items
Send the revised contract to legalNEEDS YOU
Requesting permission to send mail on your behalf.
ApproveReview diff
Summarise 38 unread threadsDONE
agi-cli — fleet
$ agents run --fleet 100 --model any --harness claude,codex,gemini
→ leased 100 workers · pinned versions · isolated worktrees
✦ researcher · analysing 2,410 filingsrunning
✦ writer · drafting 18 briefs in parallelrunning
✦ reviewer · 3 refuters per claimrunning
✓ 94 done · 6 in flight · 0 failed   — 41m saved
$
Agent orchestration

AGI CLI

One operator. A hundred agents.

The meta-harness that runs your fleets — sessions, browser, secrets, versions across every model and machine. The engine under an agent-run services firm, in one install.

Runs onmacOS · LinuxHarnessesClaude · Codex · Gemini · anyComputeyours — bring your own
Install AGI CLI →
Neutral runtime · evals · audit

Prix

The neutral runtime where agents prove and improve.

Runs any harness unmodified — on any cloud, on-prem, or the client's own VPC, isolated per client and retaining nothing by default. Every run is evaluated in place; a regression triggers a fix and a re-run, so the agent your firm delivers gets better each cycle.

Neutralityany cloud · on-prem · client VPCRetentionzero by default (ZDR)Improveseval → fix → re-run
Explore Prix →
prix — console
Eval score
92%
Regressions
0
Improved
+7%
# client: acme
harness: claude-code · any model
cloud: client-vpc · on-prem ok
retention: zero · keep score only
loop: eval → fix → replay → merge
retains nothing · score + audit only · ZDR by default
From Palo Alto

Software ate the world. Now it does the work. Phoenix rises from the ashes of the bubble to build the infrastructure for the firms doing it — client approval, agent orchestration, and a neutral runtime where their agents keep improving. From Palo Alto, California.