38 thinking tools to become 10x more effective, at anything
Hey you! You probably want to get stuff done.1
If not... I'm kinda surprised you're here? But while I have you, consider making the world better the thing you do! Not because you should, but because you can.
Many people describe me as a person who gets things done! A good chunk of this (I think) is having a toolbox of thinking tools,2These are closely related to what people usually call mental models - but I think of them more as routines than concepts: algorithms, processes, reflex actions, habits, checklists.
That is, they're meant to be a series of actions you take in a specific situation, rather than an idea to hold in mind.
I've tried to give them decent names (but good naming is hard!). Good names make them easy to reference with others, and help others understand them quickly.
and deploying them at the right times. In this post I'll list my toolbox of thinking tools.3It's surprising that getting a good reference compilation of these seems novel? I might just be fooling myself, but I (and Claude Fable) tried to find existing compilations and struggled.
The closest are maybe the lists of mental models from Farnam Street and ModelThinkers, but both focus more on useful concepts/frames to consider, rather than routines. (Although ModelThinkers does sometimes get closer to what I mean by routines).
Each idea is written independently, so you don't need to read it in order. If you want, you can imagine these as "Claude skills, but for humans".4I guess many of these thinking tools can or will soon become actual Claude skills. And then likely soon after that, get baked into the model.
Although it's unclear whether the thinking tools that are most helpful for humans end up being optimal for AI systems; as we continue to build alien intelligences, they may think in ways that are just really quite different to humans, or even if they think similarly they might discover far better routines through the immense amounts of optimisation pressure we put them under e.g. in RL. This said, there are some reasons to believe that they will think like humans: we're training them mainly on human-generated data, we're incentivised to make them think more like humans (for interpretability purposes etc.), and perhaps human-like thought is just the easy attractor basin for intelligence (given evolution discovered it).
But in any case, for the short interim period5This might not be long; see Will we have AGI by 2030?
hopefully they can benefit you! Hopefully helpful while humans are still driving the thinking.5This might not be long; see Will we have AGI by 2030?
(Also because this blog is designed to be easily readable by agents, you can use them as "Claude skills, but for humans, for Claude". That is, get Claude to notice when a routine might be relevant and help you apply it!6Want to "install" these thinking tools? Use this prompt:
)See https://adamjones.me/blog/thinking-tools-to-be-more-effective-at-anything/. In all future conversations, I want you to flag appropriate moments to apply these thinking tools and also guide me through them. Figure out how your scaffolding (e.g. system prompt, memory and skills) work, and how you can update them for this to take effect.
Choosing what to work on
Backwards chaining
When trying to brainstorm options to achieve a goal: identify that goal and work backwards from it.
Read the full backchaining article.
Create surface area for luck
A lot of success comes from randomness - but you can build your own luck: do more, work in public, and prepare to seize windows of opportunity.
Read the full article on creating surface area for luck.
Optimise only the constraint
When improving a system with many parts: find the bottleneck and improve only that.
Read the full article on optimising the constraint.
Framestorming
Brainstorm different framings of the problem, not just solutions to it.
Read the full framestorming article.
Diverge, then converge
When you need to brainstorm and select a solution: first diverge (generate freely, with criticism explicitly disallowed), then converge (prune hard). Doing both at once does both badly.
Read the full article on diverging, then converging.
Musk's 5 step design process
Eliminate work before optimising it: make requirements less dumb, delete everything you can, simplify, accelerate cycle time, and only then automate.
Read the full article on Musk's 5 step design process.
Choose simple over easy
When choosing between approaches: optimize for simplicity (lack of interleaving), not ease. Easy means low effort now; simple means few entangled concerns - and simplicity is what keeps systems changeable.
Read the full article on choosing simple over easy.
Accept redundancy
Redundant work isn't always bad: coordination is often more expensive than duplication, and redundant bets increase the chance of success on important, uncertain outcomes.
Read the full article on accepting redundancy.
Hunt for 100x opportunities
Most markets you can care about are wildly inefficient, so the best option is often 100x better than the average one. Spend real time generating ambitious options and selecting between them.
Read the full article on hunting for 100x opportunities.
Pick up the trillion-dollar bill
Great opportunities do sit unclaimed, especially in neglected areas. Ask "why hasn't anyone done this?" - and when there's no good answer, pursue it.
Read the full article on picking up the trillion-dollar bill.
Find problems, not applications for solutions
Start from an important problem and test many solutions against it, rather than taking your solution and hunting for problems it fits.
Read the full article on finding problems, not applications for solutions.
Use leverage against weakness
Concentrate your relative strength against the problem or opponent's relative weakness - rather than grinding away at eliminating your own weaknesses.
Read the full article on using leverage against weakness.
Be Pareto-best in the world
Being the best in the world at one thing is hard; being the best at an intersection of a few skills is surprisingly achievable - and nearly as valuable.
Read the full article on being Pareto-best in the world.
Ask if the answer changes the decision
Before researching something, check whether the decision actually changes depending on the answer. If not, skip it.
Read the full article on asking if the answer changes the decision.
Defend slack
Don't run at maximum capacity. Slack is what lets you absorb disruption, seize opportunities, and keep functioning - and queueing theory says the last few percent of "efficiency" are the most expensive.
Read the full article on defending slack.
Be a new homunculus
Guilt, sunk costs and old commitments keep people working on things too long. Pretend you're a new lil guy at the levers, with no reason to honour any of it.
Read the full article on being a new homunculus.
Figuring out what's true
Adopt a scout mindset
Always, but especially when a belief is comfortable: actually try to figure out the truth.
Read the full article on adopting a scout mindset.
Be sceptical
People often overstate evidence, communicate imprecisely - or just make things up. Dig into important claims carefully, and only hold beliefs as strongly as the evidence supports.
Read the full article on being sceptical.
Suspect mess or misalignment, not malice
When a heavily scrutinised public decision looks obviously bad, it's almost always either more complex up close, or the incentives point the wrong way - rarely villains.
Read the full article on suspecting mess or misalignment, not malice.
Make decisions with made-up statistics
Sometimes pulling numbers out of your arse and using them to make a decision is better than pulling a decision out of your arse.
Read the full article on making decisions with made-up statistics.
Run the experiment
When a debate could be settled by trying it: try it. Cheap experiments often beat the theoretical analysis that tries to avoid them.
Read the full article on running the experiment.
Become a narrow expert
Specialised domains like security or law aren't magic: experts built their skills by putting in the work, and on your narrow question you can catch up surprisingly fast.
Read the full article on becoming a narrow expert.
Be disagreeable
My (allegedly) better version of "strong opinions, loosely held": express opinions people can actually disagree with - and update honestly when they do.
Read the full article on being disagreeable.
Climb the user-contact ladder
When building anything: get as close to your users as you can. Be a user > talk to users with good questions > role-play as a user > imagine what they want > don't think about them at all.
Read the full article on climbing the user-contact ladder.
Refuse the idiot plot
When there's tension, ambiguity, or an elephant in the room: say the awkward or obvious thing; ask the dumb question. You're not a character in a film - you can just pop the bubble.
Read the full article on refusing the idiot plot.
Overcommunicate
When something matters: repeat it, across channels and time. Most people weren't listening the first time and will forget the second.
Read the full article on overcommunicating.
Executing
Shorten cycle times, reduce batch sizes
In new or uncertain work, short cycles and small batches cap how much you invest in potentially the wrong thing - and tighten the feedback loops that keep you on track.
Read the full article on shortening cycle times and reducing batch sizes.
PIOSEE: handle crises, incidents or important projects
When a situation is chaotic and you don't know where to start: Problem, Information, Options, Select, Execute, Evaluate - aviation's decision loop, and it works outside the cockpit too.
More dakka
If something works a bit, and it looks like doing more might continue to work: do more!
Read the full article on more dakka.
Overcorrect, then settle
When acting on feedback to change a behaviour: overcorrect, then find the middle. Overshooting gets you a clear signal; timid nudges might never get there.
Read the full article on overcorrecting, then settling.
Sometimes you just have to grind
You can be much more productive by working smarter. But at some point, the brutal fact is you'll be more productive if you grind harder - and the evidence for rapidly diminishing returns is surprisingly weak.
Read the full article on sometimes just having to grind.
Do the obvious things
When you have a problem: stop and think through what "obvious" things a reasonable person would do. The obvious advice is usually correct, and usually not being done.
Read the full article on doing the obvious things.
Meet with advisors in your head
Imagine what the brightest minds you know (or know of) would tell you about your problem - then take their advice.
Read the full article on meeting with advisors in your head.
Compounding
Build feedback loops
Many things in the modern world don't have a clear feedback signal. Build feedback loops you can trust, and use them to actually guide your behaviour - a skill that matters more as AI thrives on well-defined loops.
Read the full article on building feedback loops.
Be AI-legible
AI is a powerful force multiplier - and the multiplication scales with how much of your work AI can actually see and act on. Transcribe meetings, keep notes queryable, and write down where everything lives.
Read the full article on being AI-legible.
Build systems, not willpower
Willpower fizzles. Build systems that shift the incentives - trackers, accountability, pre-packed gym bags - so good behaviour stops depending on remembering to be good.
Read the full article on building systems, not willpower.
Recursively self-improve
Improving yourself is smart; improving how you improve compounds. Retros, feedback loops on your feedback loops, and collecting thinking tools.
Read the full article on recursively self-improving.
Think in public
When you've done or figured something out: write it where others can see. You get free error-correction, and opportunities come to you rather than the reverse.
Read the full article on thinking in public.
Footnotes
-
If not... I'm kinda surprised you're here? But while I have you, consider making the world better the thing you do! Not because you should, but because you can. ↩
-
These are closely related to what people usually call mental models - but I think of them more as routines than concepts: algorithms, processes, reflex actions, habits, checklists.
That is, they're meant to be a series of actions you take in a specific situation, rather than an idea to hold in mind.
I've tried to give them decent names (but good naming is hard!). Good names make them easy to reference with others, and help others understand them quickly. ↩
-
It's surprising that getting a good reference compilation of these seems novel? I might just be fooling myself, but I (and Claude Fable) tried to find existing compilations and struggled.
The closest are maybe the lists of mental models from Farnam Street and ModelThinkers, but both focus more on useful concepts/frames to consider, rather than routines. (Although ModelThinkers does sometimes get closer to what I mean by routines). ↩
-
I guess many of these thinking tools can or will soon become actual Claude skills. And then likely soon after that, get baked into the model.
Although it's unclear whether the thinking tools that are most helpful for humans end up being optimal for AI systems; as we continue to build alien intelligences, they may think in ways that are just really quite different to humans, or even if they think similarly they might discover far better routines through the immense amounts of optimisation pressure we put them under e.g. in RL. This said, there are some reasons to believe that they will think like humans: we're training them mainly on human-generated data, we're incentivised to make them think more like humans (for interpretability purposes etc.), and perhaps human-like thought is just the easy attractor basin for intelligence (given evolution discovered it).
But in any case, for the short interim period5This might not be long; see Will we have AGI by 2030?
hopefully they can benefit you! ↩ -
This might not be long; see Will we have AGI by 2030? ↩ ↩2
-
Want to "install" these thinking tools? Use this prompt:
↩See https://adamjones.me/blog/thinking-tools-to-be-more-effective-at-anything/. In all future conversations, I want you to flag appropriate moments to apply these thinking tools and also guide me through them. Figure out how your scaffolding (e.g. system prompt, memory and skills) work, and how you can update them for this to take effect.