BriefTechNews
Agents now outnumber humans online - and they're buying

Agents now outnumber humans online - and they're buying

7 min read · 15 sources

TL;DR
  • AI agents now generate the majority of web traffic, over half of new articles, and most production code, per a16z.
  • Shopify orders from AI-powered searches grew nearly thirteenfold year-over-year in Q1 2026.
  • Applications and neoclouds capture 68% of gross margin in the AI value chain, with apps at 40%.
  • SaaS renewal price increases average 8-12% annually, with aggressive vendors hitting 15-25%.
  • Google's Gemini Decisions API processes up to 1,000 questions in one pass at $0.10 per million input tokens.

An a16z team asked ChatGPT to make a video ad, and the agent quietly pulled in Krea, Descript, and ElevenLabs on its own - then talked the humans into a paid Descript plan to finish the job. That’s the whole thesis of the firm’s latest piece, and it’s a bigger deal than a funny anecdote about an agent upselling its own user.

The numbers behind it are the kind that make you re-read the paragraph. Agents now make up the majority of web traffic, write over half of new articles, and generate most production code, per a16z. Shopify says orders from AI-powered searches grew nearly thirteenfold in a year, and Human Security measured agentic browser traffic growing 8x between August and October. The web’s biggest customers are no longer people with eyeballs - they’re software with API keys.

Shopify orders from AI-powered searches grew nearly thirteenfold in a year, and agentic browser traffic grew 8x between August and October.

The B2A economy: agents are the new customers

The shift from B2C to B2A - business-to-agent - is about distribution mechanics, not philosophy. Retailers once blocked bots outright; now they compete for their business. The a16z piece argues that companies treating agents as first-class customers will grow, while those blocking them get left behind.

For engineers, this changes what “optimization” means. Human-facing SEO is about clicks, dwell time, and conversion funnels. Agent-facing optimization is about structured data, clean APIs, predictable response formats, and being the thing an agent can reliably call. If agents are writing most production code and buying software on behalf of humans, then your API docs are your marketing site, and your rate limits are your sales pitch. The teams that figure out how to make their product the one an agent reaches for first are the ones that win the next decade of distribution.

Who captures the margin in AI: apps win, power loses

Tom Tunguz broke down the AI value chain into five layers and normalized everything to dollars per watt per year. The gross profit pool splits like this: 3% to power, 5% to data center operators, 28% to neoclouds (before GPU depreciation), 24% to inference vendors, and 40% to AI applications.

The interesting part is what’s testing those margins. Compute prices are rising - Anthropic’s burst compute runs at $13/GPU-hour, and CoreWeave’s new contracts are signing at $20-25/w/yr. That’s pricing power, but it also means the 28% neocloud margin is under pressure from depreciation. Meanwhile, application margins vary wildly depending on whether the AI actually drives ROI for the customer.

The consolidation play is already visible: CoreWeave’s acquisitions show vertical integration is how you absorb price swings. For anyone building on top of these layers, the takeaway is that your cost structure is someone else’s pricing decision. If you’re an AI application, your gross margin depends on how much of that 40% you can defend against the neoclouds moving up-stack.

SaaS renewals: 317% price hikes and the 49% negotiation discount

Cledara’s 2026 benchmarks show SaaS renewal increases averaging 8-12% annually, with aggressive vendors pushing 15-25%. Since 2022, published list prices have risen from about 5% (Salesforce, HubSpot) to 317% for Canva Teams’ old five-person plan. Most vendors auto-renew by default and require 15-60 days’ notice to cancel.

The number that actually matters: negotiating 90 days before renewal saves 49% on average, versus 19% when you start 30-90 days out. That’s a 30-point swing for the price of calendar discipline. For anyone running software budgets, this is the argument for a renewal calendar that flags every contract at the 120-day mark. The vendors know this - that’s why they auto-renew and bury cancellation windows in legalese. The engineering answer is to treat contract management like infrastructure: automate the reminders, standardize the process, and never let a renewal sneak up on you.

The two-person GTM stack that books 10-14 meetings a month

Source: dolv.work ↗

The dolv.work piece describes a B2B SaaS team running full go-to-market with two people: publishing 4-6 pieces monthly, shipping 150 qualified first touches weekly, and booking 10-14 meetings monthly. That work previously needed a content freelancer, an SDR, and an ops contractor.

The architecture is what makes it work. Humans own judgment and external sends; agents handle research, enrichment, scoring, and drafting. Everything is grounded in the same knowledge base so outputs stay consistent, and there’s a strict review step where humans must approve or kill outputs in under two minutes. Total cost: $1,412 monthly.

The pattern is worth stealing even if you have a bigger team: the bottleneck isn’t volume, it’s quality control. Agents can generate infinite touches, but the humans who approve them are the real funnel. The two-minute review rule is the discipline that keeps the whole thing from turning into spam.

incident.io's autonomous ad agency: a Claude skill that runs campaigns

Source: growthunhinged.com ↗

The growthunhinged piece details five GTM plays, and the standout is incident.io’s autonomous ad agency built as a Claude skill in Claude Cowork. It connects to LinkedIn, Google, Meta, and X via APIs or MCP, and can create, manage, and optimize campaigns autonomously - with weekly self-improvement via pull requests.

The concrete result: marketers at incident.io can launch ads and a landing page for one target account in three minutes, a job that took their agency two or three days. Clay pointed a similar agent at attribution and found a $200,000 deal credited to a Google ad actually started two months earlier with a free trial.

The attribution finding is the sleeper insight. If agents are running campaigns and attributing revenue, they’re also exposing where the old attribution models were lying. That’s not just a cost-saving play - it’s a data-quality play that changes which channels get budget.

Gemini Decisions API: 1,000 questions, one pass, 250ms

Source: blog.google ↗

Google’s Gemini Decisions API is a new endpoint for closed-set decisions powered by the Gemini Decision 1 model. It processes multimodal inputs - text, images, audio, video, PDFs - up to 256K tokens with up to 1,000 questions, scoring options in a single forward pass. End-to-end latency is 150-250ms, at $0.10 per 1M input tokens and zero output tokens billed.

That’s nearly 5x cheaper than Gemini 3.5 Flash-Lite with the Interactions API. For routing, policy checks, and grading tasks, this is the first API that makes “ask the model 1,000 questions” a viable production pattern. The confidence scores matter too - you can set thresholds and fall back to human review when the model isn’t sure. For anyone building decision systems, this is worth a spike: the cost curve just moved.

Grok Bot 1.0: from personal assistant to team tool

Grok Bot 1.0, announced November 2, adds Team Bots for publishing and sharing, Slack and Microsoft Teams support, meetings on Google Meet, Zoom, and Microsoft Teams, shared connectors, and reviewed learnings. It’s now a team assistant rather than a personal one, with draft-based Team Bots, review queues for learnings, and app-specific shared connectors.

The 1.0 version is on Team, Enterprise, and Ultra plans, with meetings included. For engineers, the interesting bit is the review queue for learnings - that’s the governance mechanism that keeps a team-wide AI from learning bad habits. Without it, a shared bot accumulates everyone’s mistakes. With it, you get a knowledge base that improves under human review.

Ro's refounding: from zombie to $3B+ run rate

Ro was a “zombie company” in 2021: $270M run rate, 9% YoY growth. The 2022 reset cut more than $50 million in direct-to-consumer revenue because the unit economics were upside down. By 2026, Ro is at a $3B+ run rate with over 100% CAGR for three years - nearly 10x growth with fewer people (406 vs 596 peak in May 2022), and ARR per employee up more than 10-fold. It now serves more than 1% of the US population and is the leading GLP-1 provider.

The operational leverage is the story. Growth came from focus, not headcount. For founders, the lesson is that cutting revenue to fix unit economics can be the fastest path to real scale - if the underlying demand is there.

Quick hits

Source: scifivc.substack.com ↗

Fintech 3.0 is the thesis that AI agents moving money will break incumbents that rely on friction. The CFPB reported $17B in credit card late fees in 2024 - agents could eliminate that by never missing a payment. For B2B, agents handling integrations and reconciliations lower switching costs, opening AP/AR and treasury to new entrants.

Company.md is the new fundraising artifact: a markdown file written for AI consumption alongside decks and data rooms. One VC calls it a positive signal - and warns they’ll ask their AI how biased it was. Objectivity is the requirement.

Flo gets 40% of its subscriptions more than six months after install, keeping cycle tracking free and converting users when needs evolve. The argument: paywall conversion isn’t the monetization strategy; life-stage needs are.

Branding matters more when 562,579 new LLCs launched in September alone. Fewer than 15% of operators have a documented brand system before generating AI content - and it shows. GrowthPair’s $20k brand system helped reach its first $1M in annual revenue.

Get the brief

Liked this one? Pick the beats you care about and get just those, five minutes a morning. Free, no hype, one click to leave.

About Me Author

My name is

BriefTechNews

A daily digest of what actually moved in AI, tech, crypto and fintech, assembled and written with AI, with every item linked to its source. Read More
Tags

You May Also Like