My /afk-loop and auto review job are starting to pay off. The merged PRs has been double MoM.
Learning from pstack, the toolings I setup for agent to start/test/stop the work correctly has massively improve consistency and token efficiency. I don't have to babysit agents the way around my codebase anymore, but instead focusing on grilling feature and perf.
The biggest issue in agentic development by far right now is task tracking and observability into ongoing work
No task management system has ever been built to support the properties of agentic building
Properties
- realistically 10-100 tasks happening at once
- a huge range between small ones
-off tasks and large projects
- range between open-ended research and specific implementations
- multiple stages of planning, executing, review, and human approval
- the mix between needing human in the loop and not
No existing pattern solves for this
- the chat thread, even across many bots
- the kanban board (linear, notion etc)
- github and PR based tracking
- a top-level orchestrator bot that owns everything
How do we solve this???
Gemini 4 Neon is rolling out now.
Sitting between Carbon and Hydrogen, it brings frontier intelligence and fast task completion for complex software engineering, knowledge work, and science.
Cannot use Fable 5.2 as daily driver.
Either I use it wrong, or it does not fit for my kind of work.
The writing is hard to read, inconsistent, and in many cases incomprehensible. It's also lengthy and mixed up context a lot even though it "knows" the context would be used in certain order.
In some way, I agree to this view. At the same time, I wonder if it's just the same as saying using calculator will degrade our brain ability to do math.
Why they build this?
I remember Anthropic has a clause that you are not allowed using Claude to build competing service with A\. But apparently it can use to build competing service with yours.
Everyone is already doing this, I know. But a big lab enters the scene makes it discouraging for smaller players.
An effect of this is it forces others to come up with new types of product that can't be easily replicate, or not even existing yet.
I'm not in this business. Nor do I think A\ intended that effect. So I'm really curious to see how things move from here.
Starting to roll out today: Claude Whiteboards, Claude Sheets and Claude Data.
Think a plan through on a board, build a tracker in a sheet, and tell a story with your data.
For any product with a startup popup asking users to agree doing something, what's your actual KPI?
I suspect none of apps out there really care if the popup might show up right when user is tapping/clicking something and it enrolled them in immediatly.
Grok @bot app is one.
5 Habits That Are Killing Your Productivity:
1. You don’t wake up early (Marcus Aurelius)
2. You focus on what’s outside your control (Epictetus)
3. You don’t know how to say “no” (Seneca)
4. You’re in the wrong crowd (Marcus Aurelius)
5. You think you’ll live forever (Seneca)
@chrisalbon Claude Code + Codex, yes. I run both in herdr. Opus 5.5 writes most of the code, and Sol 6.1 does local review rounds before each PR goes up.