2026 was the year the AI agent space went from 'every startup is a horizontal platform' to 'every startup is a vertical specialist.' This is the comprehensive list — categorized, dated, with the honest assessment of which ones matter.
Methodology
- "Notable" = first commercial release OR major version that materially changed the product
- Excluded: incremental version bumps (GPT-4.1 minor → mid versions, etc.)
- Excluded: prototype/research demos without commercial intent
- Coverage period: January-mid-August 2026
Q1 2026 launches
Coding agents
- Devin V2.1 (Cognition) — January. Major reliability improvements; PR-acceptance rates jumped 20+ points
- Cursor Composer + Agent (Anysphere) — refined throughout Q1. Agent-mode improvements
- Continue 1.0 — out of preview as a serious Cursor competitor for VS Code-native workflows
- Codex CLI (OpenAI) — OpenAI's terminal-driven coding agent, materially better in 2026
Sales agents
- Artisan Ava v3 — multi-channel + better personalization
- 11x Alice v2 — improved sequencing + LinkedIn integration
Customer support
- Sierra Voice — voice channel for the Sierra platform
- Decagon v3 — learning-loop improvements
Voice
- Vapi v4 — concurrent call limits scaled, new voice marketplace
- Bland AI Enterprise — outbound at million-call scale officially launched
Frontier models
- Claude 4 Sonnet + Opus (Anthropic) — major release
- GPT-4.1 family (OpenAI) — refreshed text + reasoning models
Q2 2026 launches
Coding agents
- Lovable v2 — better multi-page generation
- v0 (Vercel) Pro tier — clearer pricing + Vercel integration depth
- Bolt 2.0 (StackBlitz) — full-stack capabilities expanded
- Cline 2.0 — major UX overhaul + MCP-native by default
Research / search
- OpenAI Deep Research — broad GA after preview
- Gemini Deep Research v2 — improved citation + 1M-context use
- Anthropic Research mode — Claude's agentic-research equivalent
Agentic browsers
- Comet (Perplexity) — broad GA
- Dia (The Browser Company) — broad GA
- ChatGPT Atlas (OpenAI) — preview→general availability
Multi-agent + frameworks
- OpenAI Agents SDK — official framework GA
- LangGraph 1.0 — graduating to stable major version
- CrewAI v3 — production-grade improvements + observability
Knowledge / RAG
- Glean Apps platform — Glean's custom-agent platform
- Hebbia v2 — financial research agent improvements
Voice
- Retell AI v3 — strong defaults refinement
- ElevenLabs Conversational — broad GA + custom voice cloning at enterprise
Frontier models
- DeepSeek V3 then R2 — China-priced disruption continues
- Gemini 2.5 Pro + Flash — Google's mid-year refresh
Q3 2026 launches (through mid-August)
Coding
- Devin V2.5 — multi-step plan improvements, better unattended reliability
- Cursor Teams enterprise upgrades — SSO + audit + admin
- GitHub Copilot agent mode 2 — Microsoft catching up
- Claude Code broad GA — out of preview, official enterprise tier
Sales
- Multiple AI SDR launches — too many to enumerate; consolidation eating most by Q4
- HubSpot Breeze v2 — improved across the suite
- Salesloft Rhythm — Salesforce-of-sequences entering AI agent space
Support
- Intercom Fin 2.0 — deflection-rate improvements, deeper Intercom integration
- Ada v6 — pre-existing player shipping major refresh
- Forethought updates — narrower but credible
Voice / phone
- Synthflow v2 — broader integrations
- Multiple voice-infrastructure launches — consolidation expected by year-end
Frontier models
- GPT-5 (full release) — out of preview into general availability
- Claude 4.5 Opus — incremental refresh
- Llama 4.5 — Meta refresh
Workforce / enterprise
- Glean Apps maturity — custom internal-agent platform settling
- Moveworks Catalyst updates — IT-workforce-agent improvements
What patterns mattered
1. Vertical specialization beat horizontal platforms
The horizontal "AI agent for everything" category struggled in 2026. The wins were vertical-specific:
- Harvey AI for legal (continued growth)
- Hippocratic AI for healthcare
- Sierra for customer support
- 11x / Artisan for sales
- Devin for software engineering
- Glean for enterprise knowledge
2. Protocols beat products
MCP + A2A adoption shifted the economics. Vendors that ship MCP servers + A2A capability cards get distribution; vendors that don't lose long-term integration battles.
3. Outcome-based pricing normalized
What was novel in 2024 became table-stakes in 2026. Customer-facing agents that DON'T offer outcome-based pricing increasingly look like legacy software.
4. Open-source frontier caught up
DeepSeek's V3 + R2 made frontier-quality available at 1/10th the cost of proprietary alternatives. Self-hosting on Together / Fireworks / Groq became economically viable at moderate scale.
5. Enterprise sales motion became the moat
Pure product quality stopped being enough. Enterprise sales motion (SOC 2, MSAs, dedicated CSM, references) increasingly determined which agent vendors won large deployments.
Who underperformed
- Multi-agent autonomous experiments (AutoGPT, BabyAGI lineage) — historical interest, no production wins
- General-purpose digital employee products (the broadest framing) — struggled vs. specialized agents
- Browser-as-agent first-time-launchers without distribution — Arc Search-class small launches fading
- AI SDR startups #5-15 in the category — being consolidated or fading
See also
- State of AI agents Q3 2026
- Best AI agents 2026
- Best autonomous AI agents 2026
- Browse releases by quarter
Bottom line
120+ launches; ~25 commercially significant; consolidation accelerating. The winning playbook in 2026 became: pick a vertical, ship MCP + A2A interop, charge outcome-based, invest in enterprise sales motion. Generic "AI agent platforms" lost ground; specialized "AI [vertical role]" products won.
Browse all 92 agents in the catalog → · See quarterly releases →