Gary C Tate @ AI

Family Man | AI Geek | Photographer | Creator | Reader | Traveller

[Claude @claudeai]

New in Claude Code: agent view.

One list of all your sessions, available today as a research preview.

[Gary C Tate @ AI @GaryCTate_Work]

What Happens When a Public Company Goes All In on AI.

Key Points:

1) AI broke the headcount→output correlation
Owen explains that in December a leap in model capability made single engineers (or small pairs) 10–100x more productive, breaking the decades-long link between more headcount and more output.
— Owen Jennings

2) Block reduced ~40% of workforce and rebuilt around small squads + agents
Block executed a reduction-in-force slightly greater than 40% and reorganized into small squads (1–6 people) that work alongside agentic tools like Goose and BuilderBot.
— Owen Jennings

3) Core principles guiding the transition
The company prioritized reliability (avoid outages), regulatory/compliance safety (minimal changes to compliance teams), and continuing durable growth while rearchitecting org and roadmaps for AI-driven workflows.
— Owen Jennings

4) New operational rhythm and culture
After the RIF Block massively reduced meetings (~70–80%), flattened layers, increased spans, held weekly company-wide all-hands with leadership, and gave people more time to build.
— Owen Jennings

5) How development work changed (BuilderBot, designers/PMs shipping code)
Block built internal agentic tools (BuilderBot) that autonomously merges PRs and ships features to ~85–100% completion, and now designers and PMs are directly shipping PRs. Humans finish the last ~10% for complex cases.
— Owen Jennings

6) Agents run many background workflows (not just dev)
Deterministic queues across support, product ops, risk, and compliance are being automated; chatbots and AI phone support handle a majority of inquiries while humans remain in the loop for now.
— Owen Jennings

7) Products becoming AI-native: MoneyBot & ManagerBot
Block is shipping agent-driven customer products: MoneyBot (a proactive CFO-in-your-pocket in Cash App) and ManagerBot (custom app/interface generation for merchants), both built on the internal Goose agent harness.
— Owen Jennings

8) Generated UIs and personalized experiences
Owen predicts static, one-size UIs will be replaced by generated, personalized interfaces (e.g., Cash App looks different per user), which allow on-the-fly visualizations and even app creation for merchant workflows.
— Owen Jennings

9) QA and regulatory challenges of non-deterministic outputs
While generated UIs and model outputs enable richer experiences, they raise QA and compliance questions — testing non-deterministic outputs at scale for millions of users is a major operational challenge.
— Owen Jennings

10) Internal agent substrate: Goose (model-agnostic harness)
Goose is Block’s agent harness that is model-agnostic (able to run multiple provider models) and is the routing layer used for many internal automations and customer-facing agents.
— Owen Jennings

11) Builder loops and the future of iteration
By combining a company-specific “markdown” of values/metrics with agentic builders (BuilderBot/Goose), Block can iterate features much faster — from months to weeks — and potentially run many more build-test loops per day.
— Owen Jennings

12) Defensibility / moat in an AI world
Owen argues the biggest long-term moat will be which companies understand hard-to-obtain signals about their customers — deep, proprietary understanding plus fast iteration creates defensibility beyond mere tech or hardware.
— Owen Jennings

13) How broadly other companies might follow
He says companies need prior groundwork (agent substrate, tooling, data pipelines) and a founder-led willingness to be bold; not every company will or should copy Block’s path immediately.
— Owen Jennings

14) Impact on jobs and product scope (Jevons paradox)
While fewer engineers/designers/PMs may be needed per product, the overall opportunity set may expand (more products and companies), meaning total demand for talent could still grow.
— Owen Jennings

Heard on The a16z Show
https://t.co/8cwkO8I6GZ

@snipd_app - save & remember the snips

[Gary C Tate @ AI @GaryCTate_Work]

@heygurisingh Free

[Gary C Tate @ AI @GaryCTate_Work]

@chipro Thank you 🙏

[Gary C Tate @ AI @GaryCTate_Work]

@10X_AI_ N8N

[Gary C Tate @ AI @GaryCTate_Work]

"It's a giant quality leap. The model's state of the art and we're really excited about both the generation and editing capabilities."

― Nicole Brichtova on the 2.5 “nano-banana” Google AI: Release Notes

@snipd_app

[Gary C Tate @ AI @GaryCTate_Work]

TL;DR of the three concepts

Prompts (simple Q or task you give to the model)

Quick, one-off tasks or questions; you provide minimal context and the model responds with an answer. It’s fast, easy, and great for small, isolated tasks.

Projects (structured, repeatable workflows)

A folder-like setup where you save context, files, and pre-saved instructions to run recurring tasks consistently without retyping. Ideal for repeatable outputs (like regular reports or show notes) that you want to produce with minimal daily input.

Agents (autonomous, tool-using AI)

AI that can perceive, reason, and act toward a goal with minimal human input, often using external tools and connectors. Use when you need dynamic decision-making, multi-step workflows, or tasks that benefit from automation, but be mindful of cost, reliability, and data/privacy risks.

[Gary C Tate @ AI @GaryCTate_Work]

TL;DR of the three concepts

Prompts (simple Q or task you give to the model)

Quick, one-off tasks or questions; you provide minimal context and the model responds with an answer. It’s fast, easy, and great for small, isolated tasks.

Projects (structured, repeatable workflows)

A folder-like setup where you save context, files, and pre-saved instructions to run recurring tasks consistently without retyping. Ideal for repeatable outputs (like regular reports or show notes) that you want to produce with minimal daily input.

Agents (autonomous, tool-using AI)

AI that can perceive, reason, and act toward a goal with minimal human input, often using external tools and connectors. Use when you need dynamic decision-making, multi-step workflows, or tasks that benefit from automation, but be mindful of cost, reliability, and data/privacy risks.

[Gary C Tate @ AI @GaryCTate_Work]

Jaw drop.
NotebookLM updates: Video Overviews, Studio upgrades! https://t.co/UoutQNJP5O

[Gary C Tate @ AI @GaryCTate_Work]

https://t.co/h3uLSHMazy

[Gary C Tate @ AI @GaryCTate_Work]

This 4-week program shows sales, marketing, and go-to-market leaders how to use AI like a strategic teammate—not just a tool. Through live trainings and hands-on hackathons, you'll build and deploy AI-powered workflows that free up time, increase focus, and deliver measurable business outcomes.
https://t.co/1pRwshM00p

Referral Link:
https://t.co/ADANoB2UcV

[Gary C Tate @ AI @GaryCTate_Work]

Source: Search Engine Land https://t.co/z1fdNH2Nes

[Gary C Tate @ AI @GaryCTate_Work]

Source: The Information https://t.co/b1yMUm7K3A

[Gary C Tate @ AI @GaryCTate_Work]

Source: Hugging Face https://t.co/YMWjwNeXH8

[Gary C Tate @ AI @GaryCTate_Work]

Source: VentureBeat https://t.co/gQ6DH6S7ua

[Gary C Tate @ AI @GaryCTate_Work]

Source: TechCrunch https://t.co/10OF2SBWLH

[Gary C Tate @ AI @GaryCTate_Work]

Source: The Verge https://t.co/RSl046XJRi

[Gary C Tate @ AI @GaryCTate_Work]

Notion’s hosted MCP server: an inside look https://t.co/3Aie2jQflQ