Blog

Notes on AI agents and the AI market

Sourced data and takeaways for founders, product teams and engineers. Every number says where it comes from.

State of AI 2026 for agent builders: the same Opus 5 scored 100% on structured accounting tasks and 12.3% on simulated banking assignments where delivery was checked.

Agent engineering October 16, 2026 · 8 min read

State of AI 2026 for agent builders: 10 findings, from 100% vs 12.3% to the improvement loop

What the State of AI Report 2026 means for teams that build AI agents: completion vs partial credit, the harness, security, growth and a loop to improve agents.

Read →

15 documented prompt injection cases against AI assistants and agents, disclosed 2024–2026: Microsoft 365 Copilot, GitHub MCP, Slack AI, Salesforce Agentforce, Gemini and others.

AI security October 12, 2026 · 10 min read

Nearly every AI assistant with work data was shown open to prompt injection: 15 cases

Microsoft 365 Copilot, GitHub MCP, Slack AI, Salesforce Agentforce: a registry of 15 prompt injection cases, the three conditions behind them and a checklist before you connect an agent.

Read →

An agent that succeeds 75% of the time per run passes three runs in a row only 42% of the time (Anthropic).

Agent engineering October 10, 2026 · 11 min read

How to evaluate an AI agent before launch: 75% per run means 42% three runs in a row

AI agent evaluation in practice: pass@k vs pass^k, checking end state instead of the reply, calibrating an LLM judge, and a pre-launch checklist for your eval set.

Read →

B2B monetization survey 2026: 37% of companies use hybrid pricing, 29% have AI credits and another 33% plan to introduce them within a year.

AI economics & pricing October 10, 2026 · 10 min read

AI pricing models in 2026: seats, credits or outcomes, and why hybrid became the most common choice

How AI products charge in 2026: seat add-ons, usage credits and outcome-based pricing, with examples from Intercom, Zendesk, HubSpot, Salesforce, GitHub and Cursor, survey data on hybrid pricing and margins, and a checklist for choosing a model.

Read →

Four published results on AI and developer speed: 55.8% faster on one task, 26% more completed tasks in field trials, 24% more merged pull requests at Microsoft, 19% longer for experienced developers in METR's trial.

AI in software development October 10, 2026 · 9 min read

AI developer productivity: from −19% to +56%, what each study actually measured

AI developer productivity in 2025–2026: what METR, the Microsoft field trials, the Claude Code rollout study, GitClear, DORA and Stack Overflow measured, why they disagree, and how to measure the effect in your own team.

Read →

Two true numbers: the cost of a fixed AI performance level falls about 13× a year (Epoch AI), while the cost of running frontier models rises 3–18× a year (MIT FutureTech).

AI economics & pricing October 10, 2026 · 10 min read

The same AI quality gets 47% cheaper every quarter. Why your agent bill doesn't

Epoch AI measures a 13× yearly drop in the cost of a fixed AI level; MIT FutureTech finds frontier runs getting 3–18× more expensive. Three reasons agent bills don't follow token prices, and how to price a task instead.

Read →

Google Research, Google DeepMind and MIT: multi-agent coordination improved a decomposable task by 80.8% and degraded sequential planning by up to 70%.

Agent engineering October 10, 2026 · 11 min read

Multi-agent vs single agent: when 3–10× the tokens pays off

When to use a multi-agent system and when one agent is better: what Anthropic, Cognition, Google Research and the MAST study found in 2025–2026, how many agents a task needs, and a decision checklist.

Read →

HubSpot moved from $1 per conversation to $0.50 per resolved conversation; its Customer Agent resolves 65% of conversations, by HubSpot's data.

AI in product October 8, 2026 · 12 min read

AI agents in your product: how to ship, price and prove them

HubSpot bills $0.50 per resolved conversation; Attio shipped its agent in four steps. A playbook for product teams: stages, MCP, pricing, evals, oversight.

Read →

AI share of global venture funding: about 50% in 2025, 80% in Q1 2026, over 70% in Q2, 64% in Q3 2026.

Market & venture October 6, 2026 · 6 min read

64% of venture money now goes to AI. What it means if you're raising in 2026

Q3 2026 venture data: AI took $102B of $159B, 27 companies took a third of all capital, and per-seat SaaS got repriced. What founders should change before a raise.

Read →