Jiffi AI Knowledge Hub
Most AI builds die because nobody planned them. There are pathways to build fast, and pathways to build right. This is the second one: 30 free guides, the tools that run them, and a written plan you own at the end.
Built for founders, operators and product people shipping without a dev team.
No credit card. Signing up includes 24 credits: about 8 planning turns with idea8.
Most AI content leaves you with opinions and no product. Tutorials show you a toy. Courses sell you theory. Meanwhile the thing you actually wanted to build is still a note on your phone.
These are the workflows we run on real client projects, written down in order, with the exact prompts. You plan it properly first, then you build. What comes out the other end is working software, plus the written plan that got it there.
Free to read, start to finish. Each track is a complete build, not a taster.
Case studies
Anyone can show you a finished screen. The part a demo cannot fake is the record of decisions behind it, written down while changing your mind was still cheap. These are live, and those records are still legible. That is the step most AI builds skip, and the reason most AI builds stall.
A personalised morning briefing. Trawly reads the internet overnight and delivers a finite set of stories that match one specific reader, rather than an infinite feed that matches nobody.
The information problem is not scarcity, it is ratio. The good material exists and it is buried. Newsletters solve this by writing once for thousands of readers, which means the edit is a compromise for every one of them. Feeds solve it by never ending, which is not solving it. Chatbots solve it only if you already know what to ask, and the whole point is that you do not.
So the product had a hard requirement disguised as a soft one: the output has to be finite and it has to arrive without being asked for. That constraint rules out most of the obvious designs, and it needed to be settled in writing before anyone built a pipeline, because it determines the entire shape of the system.
What you walk away with
Every idea8 session ends by writing the pack down: the decisions, the trade-offs, and the questions still open. It is what your build runs on, so the PRD, ADRs and the rest have everything they need. Below is an example, in the exact shape idea8 produces. Read it. Judge the quality yourself.
Users hit friction and never tell us. The support inbox catches the loud ones, but the quiet drop-off is invisible. We want a way for a user to say "this is broken" or "this is good" in the moment, without leaving what they are doing.
The primary user is a signed-in customer on any screen of the dashboard. The secondary user is the product team, who read what comes in.
The guides get you moving. These keep you moving when the plan gets hard, the scope creeps, or you need someone to argue with at 11pm.
Describe what you want to build. idea8 asks the questions a good co-founder would, and does not let you skip the awkward ones. Try it below.
Small-group working sessions. Bring the thing you are stuck on and leave with it running. A seat is a fixed price, and every seat comes with a bundle of credits to keep planning after the session ends.
See upcoming sessionsThe toolbelt
One balance runs the tools we host, and the set keeps growing. PalateMCP is its own product, priced separately. Every tool below states what it costs. What is live is live. What is next is marked as coming, with no date attached until we can keep it.
Argues with your idea until it is worth building.
Most planning tools take your idea at face value and format it. idea8 does the opposite. It asks the hard questions, challenges the assumptions you have not noticed you are making, and drives the conversation toward something specific enough to build. When the thinking is done, it writes the planning pack: the artifacts your build runs on, so that when you follow the guides your PRD, ADRs, brief, plan and spec have everything they need. Where the conversation did not settle a question, the pack says so under Open questions rather than inventing a plausible answer. A named gap is useful. A fabricated one is a trap.
What you walk away with
Your build stops looking AI-generated.
AI builders converge. Ask three of them for a website and you get the same centred hero, the same gradient, the same rounded cards, because they are all averaging the same internet. PalateMCP is a taste layer that sits between your AI tool and that average. It grounds the build in a library of real websites that have been analysed properly: design tokens, inner-page structure, the signature moves that make a site feel considered, and explicit do and do-not rules. Your agent stops guessing what good looks like and starts working from evidence.
What you walk away with
The guides are free, permanently. You pay for live sessions or a membership, and credits come with both. Credits are what the toolbelt runs on, never the price of a seat.
A$0
A seat is a fixed price per session. Buying one grants you a bundle of credits on top, so you leave with momentum rather than a recording.
Start with the first guide. You will have something running before you need to spend a cent.
Start building10 steps
Built the Jiffi way: the arguments were had on paper first, while changing your mind was still cheap.
The planning stage was where the awkward questions got asked. What counts as a good story for this specific reader, and how would we know we got it wrong? What happens on a slow news day, does the briefing pad itself out or arrive short? A decision like that is trivial to make in a document and expensive to discover after the pipeline is built around the wrong answer.
Only once those decisions were recorded, with the reasoning and the rejected alternatives written down, did the build start. The artifacts were not a deliverable produced for a client file. They were the working documents, and they stayed useful after launch because they explain why the system is the way it is.
Artifacts produced
Trawly is live and running at heytrawly.com. It sends a personalised briefing each weekday morning, tuned to the individual reader rather than to a subscriber list.
The reason it is here as evidence is not the launch. It is that the decisions behind it are still legible. Someone can open the record and see what was chosen, what was rejected, and why. That is what separates a product that can be maintained from one that only its author understands.
Also built this way
Every AI builds the same website. PalateMCP is the design taste layer that stops that, serving AI build tools a library of real, deeply analysed sites instead of the internet average.
See PalateMCP liveArtifacts produced
One session writes the full set
Problem, users, value, and the prompt to hand your AI
What to build first, next and later, with the reasoning
Who else is there, and where the gap is
What you are betting on, and the risk
What the build will need, nothing missed
Honest evidence, open questions included
Every tool we add runs on the same balance you already hold. Your credits reach each new one the day it lands, so what a membership is worth keeps climbing while the price stays where it is. We ship them when they are right, not to hit a date.
Attacks your plan before reality does.
You wrote the PRD. You are now the worst possible person to review it. Critic takes a finished planning artifact and argues the other side: the assumption holding up the whole thing, the requirement that contradicts another one, the success measure nobody could actually collect, the alternative the ADR dismissed too quickly. It returns a ranked list of what is most likely to break and what evidence would settle each one. Not yet available.
What you walk away with
Turns planning into something a builder can start on Monday.
A PRD is not a build queue. Handover reads your artifacts and produces the thing that actually gets worked from: scoped units of work in dependency order, each with acceptance criteria specific enough to argue about, and each traced back to the decision that produced it. Works whether the builder is a person or a coding agent. Not yet available.
What you walk away with
Tells you where the build stopped matching the plan.
Decisions rot quietly. Six weeks in, the codebase has quietly done the opposite of what the ADR said, and nobody noticed because nobody was looking. Drift checks a codebase against the artifacts that specified it and reports where the two have separated: decisions silently reversed, requirements never implemented, and constraints the build has grown out of. Then it tells you which ones matter. Not yet available.
What you walk away with
A$30/month