Husband, dad of 2 boys | AI Whisperer and all things Agents and AI | Founder of https://xitter.cf/t.co/obEl8IsEIP and https://xitter.cf/t.co/0JqxGEw6rx
Texas / Remote
Joined July 2025
- Tweets985
- Following173
- Followers94
- Likes1.6K
Patrick Robinson retweeted
You can’t build a software factory unless your team first does at least some of its product engineering through a factory as opposed to local coding agents.
This means centralizing your agents into one shared configuration that the team can optimize together. Once work starts flowing through the factory infra, you’ll have a baseline to measure and improve.
Patrick Robinson retweeted
AI means a big team can’t possibly keep track of all the daily changes.
So, I suspect the ideal team size for 2027 is 1-2 people.
Any more than that and people can't keep up with the changes, so they don't understand the system they're building.
Patrick Robinson retweeted
"The first thing you need to do to make effective use of AI and autonomous AI is to improve your own process"
-Gartner
Totally agree. I keep seeing teams over-trusting AI. They lack the automated quality controls to safely move faster.
Patrick Robinson retweeted
Subagent overuse and abuse has been a problem since mid 2025.
There is a time and a place and they can be very high leverage, but the vibeslop dopamine casino + twitter hype machine has let the average user down by encouraging more and more parallelism instead of thoughtful context management
Anthropic is now recommending that companies REMOVE unnecessary subagents from their AI systems.
This is a pretty significant shift in agent architecture, with their team finding that a single agent equipped with skills can frequently outperform complex multi-agent systems, and one customer seeing roughly 2x faster execution.
The emerging rules of thumb:
→ Skills: Teach the agent HOW to do something, including procedures, expertise, and instructions.
→ Tools / MCP: Give the agent the ability to DO things through APIs, actions, and live data.
→ Subagents: Use for parallel work, isolated context, or tasks a single agent struggles with.
The interesting part is progressive disclosure. Instead of spawning specialized agents or stuffing everything into a giant prompt, one agent discovers and loads skills and tools as needed.
Their advice is to fold existing subagents into the main loop, turn their instructions into skills, expose their actions as tools, and run evals to see if performance holds up.
As models have gotten smarter, improving agent architecture is becoming less about what you can add and more about what you can delete.
claude.com/resources/article…
Patrick Robinson retweeted
Mistake: Creating a plan via AI that answers many ambiguities in the ticket.
Instead: Update the ticket with all decisions made in the plan.
This way, the ticket is a *permanent*, centralized, comprehensive record of *all* decisions that's easy to reference later.
Patrick Robinson retweeted
"Smaller things get through the pipe faster." - Plumber
Good advice for devs too.
Patrick Robinson retweeted
If you track software factory metrics, don't celebrate the wrong signals.
Bad: LOC written, tokens used
Good: Cost-per-PR, cycle time from conversation start -> merge
A factory leaderboard should take cost into account and champion shipped work, not token spend.
Patrick Robinson retweeted
Wonder if your tests are helping? Ask Claude.
Prompt: “Look through my past sessions. How many times did a test catch a bug?”
I just asked Claude and it found 5 recent instances where a test caught a bug.
Patrick Robinson retweeted
I’ve seen many teams struggle because their tickets are too big.
I’ve never seen a team struggle because their tickets are too small.
Patrick Robinson retweeted
This post outlines an interesting new approach that AI unlocks: Non-code automated tests.
For example, QA doesn’t typically have the tech chops to write and maintain code-based tests.
But they can certainly ask Claude to click around for them!
Before you launch anything, have Opus 5.5 send 5 subagents in as your first users.
Paste this into Claude Code:
"Act as my launch QA lead. Spin up 5 subagents, each a different first-time user:
1. The impatient one: skims, clicks the first button, quits at the first error
2. The skeptic: reads pricing, terms and privacy before signing up
3. The mobile user: does everything on a 375px screen on slow 3G
4. The non-native English speaker: only follows what the UI literally says
5. The breaker: pastes junk into forms, hits back mid-flow, double-clicks everything
Each one opens the app in a real browser, goes from landing page to first success, and screenshots every moment they get confused, stuck or annoyed.
Merge everything into one list ranked by how many users hit it. Fix it all in one pass, then rerun the 5 users to prove it's fixed."
Your first 100 real users won't tell you why they left. these 5 will.
(Works best with a browser MCP connected, Playwright or Chrome DevTools)
Patrick Robinson retweeted
Still doing human code reviews? Then focused PRs are more important than ever.
If the PR contains many unrelated changes, then for every line I have to first determine what the change is for.
If the PR is focused on *one* thing, I know the general intent of each line before I even start reading.
Good news: AI is good at splitting big PRs into small, focused, stacked PRs.
Patrick Robinson retweeted
With AI, it’s tempting to create big tickets because AI can handle big tasks, and it generates code so quickly.
But the benefits of small, focused tickets remain relevant in the AI era.
- Get feedback quicker.
- Avoids context bloat / compacting.
- Less chance of spending a bunch of time going in the wrong direction.
- Leads to small PRs, which are easier to review.
- I’m more likely to understand a small ticket, and to notice gaps, inconsistencies, and logical issues in the requirements.
- Small tickets lead to smaller releases, which reduces risk.
