The agent-native rebuild of the software stack: 16 pre-AI categories now have credible agent suppliers, but only coding tops 1% of human volume
Asked:
“Compile a list of products and tools from before AI / Agents which got reimagined for the new AI / Agents world. Example: web search. What other technologies have already established AI-native providers? Who is leading each specific product category (2-3 companies)? How the AI native usage volume compares to the human usage?”
This landscape covers 16 established pre-AI product categories — from web search and email to payments, telephony and pentesting — and the 40 best-supported agent-native providers behind them, drawn from 200+ evidence records as of 1 September 2026. Each category is graded by how disclosed agent volume compares with the traditional human baseline; units range from users and sessions to calls, pages and transactions, so grades are order-of-magnitude, not market share.
The landscape matrix
16 categories × leading providers, coloured by agent-vs-human volume band · hover a mark for the metric and source date
Category cards
2–3 leaders per category where the evidence supports them; thin categories are shown thin rather than padded
What counts as AI-native? A product designed around model and agent primitives — machine-oriented retrieval, durable memory, delegated credentials, tool execution, unattended browsing, or outcome-based work — not a legacy product with a chatbot bolted on. Intercom Fin and Pentera are incumbents’ AI-native products and are labeled as such; x402 is a protocol, not a vendor.
What the matrix cannot say alone
Coding is the only category with material end-user penetration: Cursor’s ~1M daily users (and 1M+ paid by early 2026) against roughly 47M developers worldwide is low-single-digit share by user count, not parity.
slashdata.coAI search is still under 1% of classic search: Perplexity’s ~780M monthly queries are about two-tenths of a percent of Google-scale volume of 5T+ searches a year; agent search APIs add volume nobody discloses.
vellum.aiAgentMail offers the rare within-provider comparison: hundreds of thousands of agent users versus tens of thousands of human users — agents already outnumber people on the agent-native platform, while its 100M+ delivered emails round to zero against ~376B human emails a day.
agentmail.toAgent payments are tiny but measurable: 165M+ cumulative x402 transactions by April 2026 and a ~75M monthly record imply a count share in the hundredths of a percent against 266B annual real-time payment transactions — transaction count, not dollar volume.
gigs.sh
Evidence table
40 shortlist rows condensed to the strongest usage metric per provider; adoption proxies flagged
| Category | Provider | Best-supported metric (with date) | Vs human baseline | Source |
|---|
Method: built from four result sets — a 40-row category/provider shortlist, 217 detailed evidence records, 150 cross-page corroborations, and 200 traditional-usage baselines — as of 2026-09-01. “Leader” means demonstrated adoption plus repeated independent coverage, not audited market share. Volumes are heterogeneous (users, sessions, calls, pages, tool calls, transactions, resolved conversations, API events); cumulative and monthly figures are labeled; where no sound denominator exists the band reads “not comparable / undisclosed”, and no percentage is invented. Funding, valuations, benchmark scores, GitHub stars, plan quotas and ARR are treated as adoption proxies, never as usage volume. Prices and weak snippets from the shortlist were replaced with stronger metrics from the detailed set where available.