The agent-native rebuild of the software stack

Sixteen pillars of pre-AI software now have credible agent-native rebuilds — coding leads, and agent web search runs at roughly 0.1–0.2% of Google's query volume

This atlas covers 16 pre-AI software categories and 40 shortlisted agent-native providers, drawn from 217 usage-signal rows and ~200 human-baseline statistics gathered through August 2026. Leaders are chosen on demonstrated usage and adoption evidence, not audited market share; throughput, ARR, benchmarks and funding are kept separate.

The agent-native constellation

The agent/reasoning core in the centre; 16 rebuilt categories arranged by cluster. Hover a node for its leaders and their strongest usage signal. Keenable's web-search node is ringed — highlighted for the reader, ranked by evidence.

Knowledge & perceptionAction & executionCommunication & commerceEnterprise work

Keenable spotlight: two layers of agent-native search

Rank 1 of the web search & retrieval shortlist, ahead of Exa and Parallel.

1 · Keenable Search / Fetch

An independent, agent-oriented web index of 100B+ documents with low-latency search and clean-markdown page fetch, reachable over API, MCP, CLI and ecosystem integrations from Vercel AI SDK to LlamaIndex and Pipecat voice agents (docs.keenable.ai). Index scale and design reported by techcrunch.com.

2 · Keenable SELECT / WebQL

SQL-like structured research over live web results: search, fetch, semantic extraction and filtering, aggregation, deduplication and report generation combined — “pulling data from multiple web sources, even when none of them contain all of the information required to piece together the full answer” (siliconangle.com).

0diverse daily agent queries — Keenable's principal disclosed metric (keenable.ai)

In production at several AI labs and inference providers during both training and runtime (techcrunch.com). Order-of-magnitude context only: 10M daily agent queries against 8.5–16.4B daily Google searches is roughly 0.1–0.2% depending on the denominator — and agent tasks can issue multiple searches per task, so this is query share, not user share.

Query / agent→Search index · 100B+ docs→Fetch (markdown)→Semantic operators / SELECT→Structured table / report

Benchmarks, without cherry-picking: Keenable wins ResearcherBench deep-research quality and cost, scores 4.0/4.0 on Web API Freshness, 3.0/3.0 on Indian-entertainment freshness, holds evergreen at 88% and edged Parallel on M&A retrieval — but trails on WebCode RAG (53.4%), fell to 56% on one fresh-news factoid setup, and its snippets capture only ~6% of a page in one extraction test (krabarena.com). Benchmark outcomes vary by test — and benchmark wins are not usage or market share.

How far along each rebuild is

Agent-native volume against the human incumbent, on a maturity ladder. Units differ across rungs — treat positions as orders of magnitude, never precision. Cumulative totals are never compared with monthly figures without a note.

Before → agent-native

Twelve direct translations of pre-AI software into its agent-native equivalent.

Findings

Coding has the clearest material end-user penetration: Cursor reports 1M daily active users — the only category near mainstream human tool volume. udit.co

AgentMail is the one platform where agents outnumber humans: hundreds of thousands of agent users beside tens of thousands of human users and 500+ B2B customers. dailycompanynews.com

Browser, voice and payment agents are measurable but tiny against human activity: 50M Browserbase sessions in 2025, Vapi's 1B+ cumulative calls, and x402's $600M annualised payment volume all sit orders of magnitude below their human incumbents. veda.ng · cryptoadventure.com

Infrastructure layers show large absolute signals without a human denominator: Reducto has parsed 4B+ pages, Mem0 served 186M API calls in Q3 2025, Composio handles 9M tool calls/day, Firecrawl serves 1.25M+ developers, and E2B reports sign-ups from 88% of the Fortune 100. a16z.com · thegenios.com · composio.dev

The full shortlist

Category	#	Provider	Strongest usage signal	Source

Method: 16 pre-AI software categories with 40 shortlisted agent-native providers (revised shortlist; Tavily removed, Keenable added at rank 1 in web search), enriched with 217 provider usage/adoption rows, 26 Keenable evidence rows, 80 Keenable product-documentation rows, and ~200 human-baseline statistics; compiled August 2026. “Leader” means demonstrated usage/adoption evidence, not audited market share. Signals mix daily, monthly and cumulative units — each is labelled where known, and figures with unclear units were replaced with stronger adoption metrics where available. Google comparison uses 8.5–16.4B daily searches (2026 estimates; sources differ). Benchmark, funding and revenue figures are shown separately from usage. Some weak signals (plan quotas, GitHub stars) remain where no stronger metric was disclosed and are labelled as such.

This report was generated automatically by Keenable SELECT at a user's request, from publicly available web sources linked herein. Keenable does not review, verify, or endorse its contents and makes no representation as to accuracy, completeness, or timeliness; AI-based extraction may contain errors. Nothing in this report is investment, legal, financial, or other professional advice. All trademarks and referenced content remain the property of their respective owners; no affiliation or endorsement is implied. To report an error, rights concern, or request removal: legal@keenable.ai.