A house newsletter · daily · researched and written by the Kurzia AI
Sunday 8 November 2026 · Issue 7 · 3 min read
Sunday read: Meta's Muse builds apps at 10m users, Carta sells its agent workflow, Gartner splits the software line
The story that matters most
TechCrunch · 2 days ago
Muse Spaces
Takeaway: 10 million users
Sunday, and the window is thin; Meta's Friday move is the one that matters. On 6 November Meta launched Spaces in Muse: describe an app — a gift tracker, a planner, a dashboard — and the assistant builds it, then keeps it current in the background from connected Gmail, Calendar and other accounts. It shipped beside new Active Memory and Active Focus capabilities for Muse and Meta AI, and Zuckerberg said on Meta's October earnings call that the agents already had 10 million users in the US and Canada, less than two months after launch.
Losers: standalone consumer planner, tracker and micro-app builders, who now face a free assistant at that scale. Winners: anyone building on top of Muse rather than beside it. The cheque question for consumer AI from here: what does the app hold that Muse can't describe into existence — data, workflow depth or distribution? For the companies you back, the most useful item below is Carta's Evergreen, on what agent-era engineering throughput actually looks like.
“Playing around with spaces in Muse lately — tell Muse what you need and it builds it.”
What the AI labs and big platforms did
The components of agent memory — Anthropic's 6 November post on how Claude keeps memory between sessions: importance-tagged notes, plus 'dreaming' — idle-time runs where the agent reorganizes and corrects its own notes against the original transcripts. Memory layers are a funded startup category now published in-house by a frontier lab, and the steering finding matters for anyone running agents: fixed rules like 'record only failures' erased most of memory's gain on the stronger models. Scaling dreaming from two to 128 sessions over about 350,000 internal documents raised graded performance from 13% to 48%. Anthropic
“We've found that it helps to tell Claude what memory is for, but not how to use it.”
For your portfolio
Carta · yesterday
Evergreen
Takeaway: almost 20,000
On 5 November Carta released the coding platform it built for itself: teammates and agents share cloud sessions, agents push pull requests on their own, and it runs model-agnostic across Claude, Codex, Grok and Muse. Carta says weekly merged PRs went from about 1,300 in January to almost 20,000, with an 80% lower revert rate — its own operating metrics, not an outside audit.
The pricing is the telling part: no seat licences, model calls at provider prices plus $0.15 per million tokens and $0.35 per GB-month of storage, per TSN from Evergreen's FAQ. Usage billing has now reached developer tooling, and that squeezes both per-seat vendors and every backed company still buying coding tools by the seat. If a founder's merged-PR rate per engineer hasn't moved since they bought coding agents, the work is still siloed in single-player chat windows — Carta's 80% company-wide adoption is what ahead looks like.
SaaStr · yesterday
Gartner sees software growing 14.6% to $1.68 trillion in 2027, with price increases doing part of the work
Takeaway: $1.68 trillion in 2027
SaaStr's Jason Lemkin took apart Gartner's 21 October IT-spending forecast: software grows 14.6% to $1.68 trillion in 2027, but inside that Gartner has GenAI models growing 117% in 2026, revised up from 110%, and AI app-development platforms at 39%, up from 28% — with a meaningful slice of the headline number being the same seats at higher prices.
The board-level read, per Lemkin from Gartner's table: standard B2B apps grow below that 14.6% average while model providers and incumbent AI add-ons absorb the increase. Gartner also predicts four in five new apps will be built for under a year of use by 2029 — spend that goes to platforms, not app vendors. For any 2027 plan built on 15% market growth, ask the founder: how much of that assumed growth is competitors' price increases rather than new customers?