Entries are representative, not ranked. Capabilities and pricing change weekly; verify before you commit budget.

How we picked these domains and products

• Live recency check: web search per domain for launches, shutdowns, and “legacy” labels—not model memory alone.
• Official sources first, then cross-check 2026 comparisons and usage snapshots (e.g. OpenRouter traffic for images).
• Same depth for every product, including K2F in the documents domain.
• As of September 2026. Not exhaustive.

What we excluded (and why)

• OpenAI Sora (consumer app and Videos API): discontinued in 2026 (app April 26; API September 24) with no documented replacement endpoint—teams moved to Veo, Kling, and Seedance for new work.
• GitHub Copilot as a 2026 “coding agent” peer: still maintained for inline completion inside GitHub, but the autonomous agent conversation in mid-2026 centers on OpenAI Codex, Cursor, and Claude Code—not 2022-era “Copilot vs everything” framing.
• Midjourney as the default image API story: still a consumer product, but developer/API traffic in 2026 heavily favors Google’s Nano Banana family and FLUX.2 for shipped pipelines.

At a glance

Coding agents

Autonomous coding in 2026 is a three-way workflow choice (terminal vs IDE vs ChatGPT bundle)—plus optional open harnesses like OpenCode if you want MIT-licensed tooling. None of these replace a document compiler.

OpenAI Codex

• OpenAI’s coding agent stack: open-source Codex CLI in the terminal, IDE extension (VS Code, Cursor, Windsurf), and Codex Web at chatgpt.com/codex (CLI docs).
• Supports local repo work, scripted codex exec, and cloud delegation (codex cloud) to run tasks remotely and apply diffs locally.
• Agent fit: High—MCP, skills ecosystem, multi-surface same agent family.
• Limitations: Plans, credits, and cloud features vary; not a document layout engine. Pair with K2F for publish gates (compare Codex).

Cursor

• Agent-first workspace (Cursor Agent); Cursor 3 emphasizes parallel local/cloud agents and broad model routing.
• Best when you want the agent inside the editor with Tab completions and visual diffs.
• Limitations: Proprietary; monthly credits by model tier.

Claude Code

• Terminal-first agent bundled with Claude subscriptions; strong for long-horizon repo tasks and hooks-driven automation.
• Best when you live in the shell and want Anthropic’s orchestration story.
• Limitations: Separate billing from ChatGPT/Codex; not multi-vendor by default.

AI video generation

After Sora’s 2026 shutdown, production teams standardize on models with active APIs and native audio. Clips remain short; plan editorial finishing separately.

Google Veo 3.1

• Google’s cinematic line; widely cited in 2026 roundups for realism, prompt adherence, and native audio (including vertical/4K modes in 3.1-era specs).
• Limitations: Product entitlements and duration caps depend on Google’s consumer/cloud surfaces.

Kling 3.0 / Omni

• Strong for multi-shot storytelling, camera control, and synchronized dialogue in recent tiers.
• Limitations: Regional and reseller pricing varies.

Seedance 2.0

• ByteDance model noted for multi-shot consistency and multimodal references in 2026 benchmarks.
• Limitations: Moderation and regional availability differ by host platform.

Runway

• Studio and API for generative video (Gen-4.x family on runway.com); still a common incumbent beside newer foundation-model hosts.
• Limitations: Credit-based pricing; clip length limits.

Audio & voice AI

ElevenLabs

• TTS, dubbing, and voice APIs (elevenlabs.io) for apps and media pipelines.
• Limitations: Licensing/consent for cloned voices; usage-based pricing.

Suno

• Prompt-to-music for creators (suno.com).
• Limitations: Commercial rights governed by product terms—not a replacement for licensed production music.

Image generation

2026 API traffic skews toward Google’s Nano Banana line and developer-deployable FLUX.2—not 2023 Discord-first defaults.

Nano Banana 2 (Gemini 3.1 Flash Image)

• Google’s fast image model (DeepMind overview); launched in the Nano Banana 2 wave (Feb 2026) with editing, consistency, and production resolutions.
• Limitations: Verify factual claims in images; watermarking via SynthID/C2PA.

FLUX.2 (Black Forest Labs)

• Developer-oriented image stack (API and open checkpoints such as FLUX.2 Klein) for cost-sensitive or on-prem generation.
• Limitations: License terms vary by checkpoint; you own hosting and safety filters.

AI research & agentic browsing

Perplexity (Comet + Computer)

• Comet browser (perplexity.ai/comet) delegates research, email drafts, and web chores from a dedicated client.
• Computer / Browser Control runs actions on sites; as of August–September 2026, heavy Computer use is credit-metered even for some Pro subscribers—budget beyond the base subscription.
• Limitations: Not a document compiler; automation costs can spike after promotional credits.

Enterprise workspace copilots

Microsoft 365 Copilot

• Copilot across Word, Excel, PowerPoint, Teams with tenant Graph context (Microsoft 365 Copilot).
• Limitations: Admin and licensing complexity; proprietary Office formats.

Google Workspace with Gemini

• Gemini in Gmail, Docs, Sheets, Slides (Workspace AI).
• Limitations: Strongest inside Google-native files.

Legal & regulated professional AI

Harvey

• Legal platform (Assistant, Vault, agents) with enterprise security (harvey.ai).
• Limitations: Vertical SaaS—not an open publishing format.

Thomson Reuters CoCounsel

• Legal research and drafting within Thomson Reuters subscriptions (CoCounsel).
• Limitations: Ecosystem-bound.

Customer support & service AI

Intercom Fin

• AI agent for conversational support on Intercom (Fin).
• Limitations: Platform-bound.

Zendesk AI

• AI agents and copilot features for ticketing (Zendesk AI).
• Limitations: Requires Zendesk estate.

Sierra and Decagon (enterprise)

• Sierra (sierra.ai) and Decagon (decagon.ai) market AI customer-service agents for large brands—often alongside—not instead of—ticketing platforms.
• Limitations: Enterprise sales cycles; verify product scope vs your CRM stack.

Documents, publishing & agent-editable deliverables

SaaS agents (Manus, Gamma) optimize speed in a hosted UI. K2F optimizes machine editability, lock-based pixels, and multi-format export with verification.

K2F

• Open semantic document format and reference engine (k2f.dev); 0.3.x ships PDF (incl. AcroForm), DOCX, PPTX, IDML, Markdown bridge, agent skill, and stdio MCP (status).
• Export accuracy: PDF matches the lock pixels; DOCX/PPTX/IDML allow slight Office/InDesign reflow and are not fillable forms (AcroForm is PDF-only).
• Agent fit: High for pack_verify gates and one source → many exports; complements Codex for code/data orchestration.
• Limitations: Exports are lock drawings, not round-trip sources (DOCX_IS_NOT_A_SOURCE, PDF_IS_NOT_A_SOURCE, etc.); not a drag-and-drop Word replacement.

Manus

• Autonomous agent for research-backed documents, slides, and Wide Research parallel tasks; API v2 for automation (Manus API).
• Limitations: Proprietary outputs; review stakeholder-ready claims manually.

Gamma

• AI decks, docs, and sites with PPT/PDF/Google Slides export (gamma.app).
• Limitations: Hosted editor; no open verify gate on semantic source.

Genspark

• Agentic workspace for research, pages, and slides (genspark.ai)—overlaps Manus/Gamma on “prompt → deliverable.”
• Limitations: Proprietary; export and API details vary by plan—verify for production.

How document publishing fits the wider AI stack

1. Coding agents (Codex, Cursor, Claude Code) — plan, implement, analyze data, call APIs.
2. Document SaaS (Manus, Gamma) — fastest path from prompt to presentable deck or report.
3. Document compiler (K2F) — semantic patches, deterministic lock, integrity banners, and exports your CI can gate.

A common 2026 pattern: Codex (or Claude Code) owns the repo; K2F owns the .K2F package and exports; Gamma still wins when marketing wants a polished deck without touching git. See Compare, Gallery, and Why K2F exists.

When to reach for what

• Default agentic coding stack → Codex, Cursor, or Claude Code (pick by surface: ChatGPT bundle vs IDE vs terminal).
• Video after Sora → Veo, Kling, or Seedance—confirm API access on your host.
• Images in production APIs → Nano Banana 2 or FLUX.2—not assumed 2023 leaders.
• Browser delegation → Perplexity Comet; monitor Computer credits.
• Verified Word/PDF/PPTX/IDML from agents → K2F skill + compiler.
• Research doc + slides in one SaaS loop → Manus, Genspark, or Gamma.

Sources & freshness

This roundup is based on publicly available information at the time of writing. Capabilities, pricing, and policies may change; verify details with the relevant provider.