Deep Work vs Multitasking, OpenAI cofounder joins Anthropic, and Gemini 3.5 Flash

CTO Circle

Reply-To: marco@pullpo.io

Issue 008

Eighth issue

Run IT on autopilot

CTOs running IT on the side know the drill: slow onboarding, leaky offboarding, untracked SaaS, no audit trail for ISO 27001 or NIS2.

Factorial IT solves all of it from one platform: laptops shipped and configured in minutes, one-click offboarding across every app, full device and SaaS inventory, and audit-ready compliance reports for ISO 27001, SOC2, and NIS2.

Across Mac, Windows, iOS and Android. Fully managed fleet in under a week.

Run IT on autopilot

Next community events.

Last CTO Lunch before summer - Barcelona - Itnig - June 17th.

CTO networking, tapas, and short talks from engineering leaders. Apply here!

Short opinionated trend: Deep work vs multitasking for developers.

Five years ago, most of my work was single-threaded. As a CS student and developer, it was mostly one task, one problem, long blocks of focus.

Then I started pullpo.io, and that changed overnight. Suddenly, my day was split between product, customers, bugs, sales, legal, support, contracts, emails, and random urgent things at the same time.

I think most developers are facing a similar shift: waiting for AI agents responses, worktrees, async reviews, CI, Slack.. make engineering less purely single-threaded.

I find it useful to be conscious of which work mode you are in:

Deep work mode. Single thread.

For architecture, complex debugging, defining specs and requirements, risky reviews, setting up the AI harness.

Orchestration mode

This is managing computers working for you: agents coding your requirements, tests running, small PRs, docs, refactors, and parallel branches.

Here, being too single-threaded can be slow and break your flow too.

Admin / communication mode

For messages, ticket clarification, quick feedback, follow-ups, and unblocking people.

Here, deep work is overkill.

The hard part is knowing when to switch based on business needs, and also how you feel.

Single-threading has a cost: waiting.

Multi-threading has a cost: context loss.

News, reads and launches.

Google launched Gemini 3.5 Flash. Gemini 3.5 Flash is Google's new fast frontier model for agentic workflows, coding, and real-world task execution.

OpenAI co-founder Andrej Karpathy joined Anthropic's pretraining team, giving Claude a major AI talent win.

Project Glasswing: what Mythos showed us. Cloudflare says Anthropic's Mythos Preview shows AI can now chain small bugs into real exploits, making security harnesses (not generic coding agents) the new defensive layer.

From Forward Deployed Engineers to Forward Deployed Software. Forward deployed engineers help AI work the first time; forward deployed software makes it work again and again without starting from scratch.

OpenAI is bringing Codex to mobile so you can start, steer, approve, and monitor long-running work from your phone while it runs securely on your computer or remote environment.

A History of IDEs at Google. Google spent years trying to make IDEs work at monorepo scale, and the big lesson is that standard tooling creates massive leverage: once most engineers used Cider V, every integration, extension, and AI feature became much more valuable.