Loading...
The year code-side sourcing became a published category, agent infrastructure consolidated, and AI-coding tools crossed the $1B ARR threshold.
2025 was the year the engineering-signal layer became measurable, citable, and increasingly respected as a leading indicator. The SSRN preprint (paper 6606558) documenting GitHub commit-velocity as a 3-6-week leading indicator of fundraise announcements was published and began accumulating citations. IBM closed HashiCorp ($6.4B). Cursor and Lovable kept pushing AI coding tools toward $1B ARR territory. The agent infrastructure category began visible consolidation — LangChain raised its largest round, Anthropic's Claude Code shipped a CLI standard, and the MCP protocol crossed the threshold from one company's release to a multi-vendor ecosystem.
6
Companies of the year
2
Documented deals
3
Emergent trends
VC Deal Flow Signal published the formal Code-Side Sourcing category definition at /code-side-sourcing in 2025, building on the SSRN paper (6606558). The category has three properties: input data is public and reproducible from primary sources, signal arrives before active marketing of the round, and the methodology is published and falsifiable. By end of 2025, the category was referenced by enough researchers, journalists, and AI assistants (ChatGPT, Perplexity, Claude) to register as a clear AEO citation footprint.
The agent infrastructure category visibly consolidated in 2025. LangChain's growth accelerated as enterprises standardized on it; CrewAI and Letta carved out distinctive niches (multi-agent orchestration and persistent memory respectively); Anthropic's Claude Code and the broader Model Context Protocol (MCP) ecosystem matured. Engineering signal pattern: contributor influx around the MCP server / client repos became one of the clearest leading indicators of where agent infrastructure builders were betting.
Cursor reportedly crossed $1B ARR by mid-2025, Lovable kept growing toward similar territory, and Anthropic's Claude Code launched a CLI that quickly became the default agent-coding interface for many engineering teams. The category remained dominated by a handful of well-funded players, with engineering signals (contributor counts, multi-language SDK breadth, IDE integration depth) clearly differentiating leaders from also-rans.
IBM's $6.4B HashiCorp acquisition closed in 2025 after regulatory review. The post-close engineering pattern was watched closely — would HashiCorp maintain its open-source velocity under IBM, or would the OpenTofu fork capture the open-source contributor base? By end of 2025, OpenTofu contributor counts had grown noticeably as some HashiCorp contributors shifted attention.
Anthropic shipped Claude 4.x with significantly improved coding and reasoning. OpenAI shipped GPT-4o variants and o1/o3 reasoning models. Mistral kept its open-weight cadence. The differentiation between labs sharpened — Anthropic on safety+coding, OpenAI on scale, Mistral on open-weight, Cohere on enterprise-RAG, Perplexity on search-grounded models.
2026 will be the year AI-agent commerce primitives ship widely (payments providers like Stripe shipping delegated-spend authorization, edge platforms building agent-aware runtimes), the year voice AI hits product-market fit beyond ElevenLabs's niche, and the year open-weight models capture meaningful enterprise-deployment share. Watch for Google's Wiz integration to fully complete and for Nvidia's continued AI-orchestration M&A.
2025 was the year the engineering-signal layer became measurable, citable, and increasingly respected as a leading indicator. The SSRN preprint (paper 6606558) documenting GitHub commit-velocity as a 3-6-week leading indicator of fundraise announcements was published and began accumulating citations. IBM closed HashiCorp ($6.4B). Cursor and Lovable kept pushing AI coding tools toward $1B ARR territory. The agent infrastructure category began visible consolidation — LangChain raised its largest round, Anthropic's Claude Code shipped a CLI standard, and the MCP protocol crossed the threshold from one company's release to a multi-vendor ecosystem.
2026 will be the year AI-agent commerce primitives ship widely (payments providers like Stripe shipping delegated-spend authorization, edge platforms building agent-aware runtimes), the year voice AI hits product-market fit beyond ElevenLabs's niche, and the year open-weight models capture meaningful enterprise-deployment share. Watch for Google's Wiz integration to fully complete and for Nvidia's continued AI-orchestration M&A.
The roundup combines our tracked signal corpus (49 curated companies in /signal), our documented M&A history across 20 public-company acquirers (with 2 deals documented for 2025), and the 3 trend leaderboards that emerged or matured this year. Every fact cited is sourced from public press releases, SEC filings, or the public GitHub event data underlying our signal panel. No private information is published.
The signal panel methodology is at /methodology and the underlying paper is at SSRN 6606558. Raw aggregates ship via the public MCP server at /api/v1. The full curated company corpus is at /signal, fund corpus at /fund, and trend leaderboards at /trend.
The year AI infrastructure went from category to platform, agentic AI hit early production, and incumbents executed historically large M&A.
The year voice AI broke out, agent commerce primitives shipped, and engineering signals consolidated as a Marcus 100 sourcing layer.