The 100%-or-Nothing Automation Rule
An automation that works 95% of the time isn't an automation — push it to 100% or don't rely on it
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 4
- Confidence
- 90%
Cat's discipline for capturing AI leverage: find the tedious tasks you do repeatedly, hand them to an AI tool, and grind the automation all the way to 100% reliability. Most people stop at 90-95% and give up, but that last stretch is what makes an automation trustworthy — and a 95% automation carries little real value because you still have to babysit it. The payoff is reclaiming time for the creative work and the pet projects you never had bandwidth for.
Origin
Cat Wu's advice for thriving in an AI-driven world; she and the host both admit being guilty of stopping short.
Core principles
- 01Target repetitive tasks you do multiple times — those are the automation candidates
- 02An automation below 100% isn't really an automation; it still demands supervision
- 03The last 5-10% takes disproportionate effort but is where the value lives
- 04Building the automation is often slower than doing the task once — invest anyway for the compounding leverage
- 05Only build apps and automations you actually use every single day
- 06Don't over-customize your setup to the point it distracts from the real work
How to run it
- 1
Spot the repeated manual task
Notice any tedious task you do multiple times and flag it as an automation candidate. Prioritize the parts of your job you hate over the creative parts you love.
- 2
Build the automation with an AI tool
Use Claude Code, Co-work, or similar to draft the automation, letting it learn from how you've done the task before.
Watch out Expect building it to be slower than doing the task manually once — that's the cost of the compounding payoff.
- 3
Grind the last 5-10% to 100%
Put in the elbow grease: teach the tool your preferences, give it feedback, and iterate until the success rate is essentially 100%. Don't give up at 90-95%.
Watch out A 95% automation that occasionally misfires (e.g. mis-filing a real email as spam) can cost more than it saves — reliability is the whole point.
- 4
Redeploy the reclaimed time
Once the automation is trustworthy, use the freed bandwidth for the creative work and the pet projects you never had time for.
Pro tip Build only automations and apps you'll actually use daily — usage is where the value is realized, not one-shot prototypes.
Watch out Beware the opposite failure: obsessively customizing your setup (piling on skills and MCPs) until it distracts from the core task you set out to do. Simple setups often work better.
In the wild
The host built a workflow that auto-categorizes cold-pitch emails into a 'spammy' folder. It works ~95% of the time but occasionally buries a real email, so he can't fully trust it — a textbook case of an automation stuck short of 100%.
→ Identified as needing the last-mile push to 100% before it can be genuinely relied on.
A Claude Code sales colleague, tired of rebuilding similar decks, built a web app seeded with the core proven decks (101, 201, Mastering Claude Code) that pulls customer context from Salesforce, Gong, and notes to auto-tailor slides (adding HIPAA or code-review slides, removing enterprise-only features as relevant).
→ Work that took 20-30 minutes manually now takes a few seconds and produces a tailored deck reliably.
Common mistakes
Stopping at 90-95% reliability
An automation that mostly works still requires supervision and can cause real errors, so it delivers little leverage. The last 5-10% is exactly what makes it trustworthy enough to rely on.
Over-customizing the setup instead of shipping
Obsessively adding skills, MCPs, and workflow tweaks can become a distraction from the actual goal of building the product or feature. Simple setups often outperform elaborate ones.
Is it for you?
Best for
Knowledge workers and builders who want durable leverage from AI on recurring tedious tasks
Not ideal for
One-off tasks where building a reliable automation costs more than ever doing the task by hand
From the transcript
“anytime you realize that you're doing some manual task multiple times, think about how you can use Claude Code, Co-work, or other AI tools to…”
“If an automation doesn't work 100% of the time, it's not really an automation. And that last 5 to 10% does take more time.”
“There There's just not much value in a 95% there automation.”
“I would really push people towards building apps that you're actually using every single day cuz I think only through that usage are you actually…”
From the episode
How Anthropic’s product team moves faster than anyone else
Cat Wu (Head of Product, Claude Code)