Cloud Coding Agent Platforms

Executive Summary

A clear category has crystallized in agentic engineering: platforms where coding agents run in managed cloud environments, take delegated tasks from Slack, Linear, tickets, or a web dashboard, and return pull requests — at parallel, fleet scale. The developer's laptop stops being the execution venue; the PR review becomes the human checkpoint.

This report was previously titled Model-Agnostic Agentic Engineering Platforms. We renamed it in June 2026 because the model-agnostic axis stopped being the story: Devin now offers OpenAI, Claude, and Gemini models; Amp routes across three labs internally; nearly everyone supports multiple providers. What still divides the market is the execution model — cloud delegation versus local supervision — and within cloud delegation, whether the platform runs its own agent (Devin, Factory) or orchestrates yours (Tembo, Blocks, 8090).

Key Findings:

• Capital validated the category this quarter — Cognition's $1B+ Series D at $26B (~$492M run-rate), Factory's $150M Series C at $1.5B, Niteshift's Greylock-backed GA on June 10^[1][2][3]
• Entry pricing collapsed — Devin went from ~$500/seat to a $20 floor; Replicas and Niteshift publish transparent per-seat tiers; the premium has moved to fleet scale and enterprise governance^[4]
• The labs compete directly now — Claude Code on the web runs parallel cloud sessions with Routines; Codex cloud tasks ship Automations and a 4M-weekly-developer funnel^[5][6]
• BYOA is the standard architecture — orchestrators wrap Claude Code/Codex subscriptions rather than reselling tokens; orchestration UX, integrations, and governance are the differentiators
• Integration breadth separates orchestrators — Tembo leads with 11 native integrations (GitHub, GitLab, Bitbucket, Linear, Jira, Sentry, PostgreSQL, Supabase, AWS, Slack, Notion); competitors average 2-4^[7]
• The enterprise extreme exists — 8090 sells fully managed software delivery from $1M/yr, with the striking term that 8090 retains codebase IP^[8]

Strategic Planning Assumptions:

• By 2027, teams will evaluate these platforms primarily on automation, integration, and governance capabilities, not model support
• By 2028, first-party cloud surfaces (Claude Code web, Codex cloud) will own the low end; independents survive on multi-agent orchestration and enterprise depth
• By 2029, "delegate to the cloud, review the PR" will be the default workflow for routine engineering work

---

Market Definition

Cloud coding agent platforms run AI coding agents in managed cloud environments and operate on a delegation model. Three criteria define membership:

Inclusion Criteria:

• Agents execute in managed cloud/sandboxed environments (not solely the developer's machine)
• Work is delegated as tasks (chat, tickets, schedules, webhooks) and returned as reviewable pull requests
• Supports parallel/fleet execution
• Commercial product or mature open-source project, actively developed

Exclusion Criteria:

• IDE-first tools where the editor is the product (Cursor qualifies only via its background agents)
• Local-first CLI agents and desktop orchestrators (OpenCode, Amp, Conductor, 20x — see AI Coding Assistants and Mac Coding Agent Apps)
• Agent-enablement layers without execution (Tessl — covered with Agentic Skills Frameworks; Entire's provenance capture — covered with Developer Trust Tools)
• Pure CI bots without general engineering scope

---

Comparison Matrix

Sorted by automation + integration depth within approach groups

---

The Three Approaches

Own-Agent Platforms (Devin, Factory, Replicas, cto.new)

The platform ships its own agent, tuned to its own harness. Strongest single-task benchmarks (Factory claims Terminal-Bench #1) and the simplest buying decision — but you adopt their agent's strengths and weaknesses wholesale.^[2][10][11]

Orchestrators / BYOA (Tembo, Blocks, Niteshift, 8090, Background Agents)

You bring the agents — Claude Code, Codex, Cursor, OpenCode — and the platform supplies cloud execution, parallelism, integrations, and governance. A hedge against agent churn; orchestration quality is the product.^{[7][12][3][13]}

First-Party Cloud Surfaces (Claude Code web, Codex cloud, Cursor background agents)

The labs' own delegation surfaces: cheapest entry, deepest model integration, weakest cross-tool workflow integration. Their existence caps what independents can charge for basic cloud execution.^[5][6][14]

---

Strategic Recommendations

By Use Case

By Buyer Profile

Individual developers: start with your lab's first-party surface — Claude Code on the web or Codex cloud — and graduate to a platform when you need cross-tool workflows.

Startups: Niteshift or Replicas for transparent per-seat pricing with BYO credentials; cto.new for free experimentation (ad-supported).

Scale-ups with team workflows in Slack/Linear/Jira: Tembo or Blocks — the orchestration plus integration depth is the product.

Enterprises: Factory or Devin for own-agent scale with governance; Ona for VPC-deployed environments; 8090 if you want delivery fully managed (and can accept its IP terms).

---

Market Outlook

Near-Term (2026)

• First-party surfaces add integrations (Codex already has Slack/Linear), compressing the orchestrators' moat — integration breadth and multi-agent support become the survival traits
• Expect at least one orchestrator acquisition by a lab or platform vendor
• Pricing stratifies: $0-20 first-party entry, $50-300/seat platforms, $1M+ managed delivery

Medium-Term (2027-2028)

• Fleet governance (budgets, audit, approval gates) becomes the enterprise procurement checklist
• The own-agent vs BYOA distinction blurs as own-agent platforms add external-agent support (Devin's multi-model turn is the leading indicator)

Long-Term (2029+)

• Cloud delegation becomes the default for routine engineering work; local supervision reserved for novel or sensitive work
• The PR as the human checkpoint either holds — or gets replaced by policy-based auto-merge with provenance trails

---

Bottom Line

The category formerly known as "model-agnostic platforms" is really the cloud delegation market, and it just got its proof points: $26B Devin, $1.5B Factory, Greylock backing a days-old GA, and both major labs shipping first-party competitors.

Own-agent leaders: Devin (scale), Factory (enterprise + benchmarks). Orchestration leaders: Tembo (integration depth, agent-agnostic), with Blocks and Niteshift pressing. The floor-setters: Claude Code on the web and Codex cloud — first-party, low-cost, and improving monthly.

The strategic question for buyers: do you want the best single agent in the cloud (Devin/Factory), your choice of agents in the cloud (Tembo/Niteshift/Blocks), or the cheapest cloud surface for the agent you already pay for (the labs)? The honest answer for most teams is that they'll use two of the three.

---

Research by Ry Walker Research • methodology

Disclosure: Author is CEO of Tembo, which competes directly in this category. This report follows the same agent-driven, source-cited process as every other report on this site.