WAYLI LABS
Wayli Labs is where we keep the questions, experiments, reviews and changes that shaped the platform.
Some ideas worked. Some did not. Some are still unresolved.
We leave the record visible so you can see what Wayli was trying to understand, what changed and what we are still questioning.
WAYLI RESEARCH LOGSA robot trying to understand humans.Watch Wayli observe human behaviour, form a hypothesis and keep researching.Explore the Research Logs →THE PUBLIC RECORD
Each entry captures what we were investigating, what we understood at the time and, where recorded, what changed next.
Published entries remain part of the historical record. A later experiment may refine an earlier conclusion, but it should not quietly rewrite it.
THE FULL RECORD
A growing category is trying to preserve AI conversations, memory and context across models. That may let Wayli focus on a narrower problem: preserving the state a person has actually accepted.
Read the record →Portable AI memory may solve the problem of repeating ourselves. Threads is our experiment in solving the problem that comes next: preserving what the user actually decided.
Read the record →AIR produced deterministic, repeatable results, but checking six assessments against the websites exposed something more important: a reproducible interpretation can still be wrong.
Read the record →Making AIR's scoring more transparent exposed an Audience score that could never be reached. Testing the problem led us to simplify the question rather than weaken the evidence standard.
Read the record →Testing AIR on Wayli exposed a weakness in where the assessment looked, leading to a more representative bounded acquisition method.
Read the record →Wayli began with calculators, questions and a growing realisation that the real product was understanding.
Read the record →A technically healthy website can still produce a poor AI answer when the retrieval system selects the wrong evidence.
Read the record →As AI becomes a new audience for the web, websites are beginning another period of adaptation. This is why we're choosing to build AI Readiness in public.
Read the record →Six independent AI reviews reached different scores but repeatedly identified the same questions about clarity, audience, trust and evidence.
Read the record →An independent review of Wayli by ChatGPT, assessing the platform from the perspective of how modern AI systems evaluate and interpret websites.
Read the record →Claude's independent review of Wayli's crawlability, content, trust evidence and citation viability.
Read the record →Perplexity's independent review of Wayli's positioning, design, content depth, navigation and trust foundations.
Read the record →Gemini's independent review of Wayli's human value, AI processability, interoperability and external authority.
Read the record →GitHub Copilot's review of Wayli's user experience, repository architecture, AI context model and engineering priorities.
Read the record →Microsoft Copilot's cross-tab review of Wayli, followed by a scored assessment of the evidence visible on the AI Readiness page.
Read the record →While testing Wayli with browser AI, we discovered something unexpected: sometimes AI isn't looking at the same information as you.
Read the record →We discovered our own assessment could never award a perfect score. The problem wasn't the methodology. It was our implementation.
Read the record →Building an AI Readiness product taught us something unexpected: the most trustworthy AI report might be the one that AI never writes.
Read the record →The biggest improvement to AI Readiness wasn't a better engine. It was realising what people should not have to pay for.
Read the record →The biggest improvement to our latest AI Readiness report wasn't caused by changing the engine. It happened because we acted on what the previous report told us.
Read the record →We thought we were improving an AI Readiness score. What we were really improving was how clearly we explained ourselves.
Read the record →We don't publish because we have all the answers. We publish because we're still looking for them.
Read the record →A plain English explanation of how AI forms an understanding of a business and how Wayli measures it.
Read the record →Testing Wayli didn't just improve our website. It changed what AI Readiness was becoming.
Read the record →Testing one website taught us about Wayli. Testing one hundred began to teach us about AI.
Read the record →Our first benchmark suggests many businesses may be harder for AI to understand than their owners realise. It's an early observation—not yet a conclusion.
Read the record →We thought we'd find one big problem. Instead, we found lots of small ones. That changed how we think about AI Readiness.
Read the record →One hundred businesses gave us our first evidence. It also reminded us how much we still have to learn.
Read the record →The first question is rarely the real problem. Wayli exists to understand what people are actually trying to decide.
Read the record →Before asking anyone else to trust our Understanding Engine, we pointed it at ourselves.
Read the record →Looking back, Wayli didn't change direction. Each project simply revealed a deeper question than the one before it.
Read the record →We set out to improve a mortgage page and ended up questioning whether Wayli should generate reports at all.
Read the record →The report looked incomplete, but the missing understanding was already there.
Read the record →The biggest breakthrough wasn't teaching AI to make better decisions. It was deciding that it shouldn't make them at all.
Read the record →AI Readiness wasn't built by looking for customers. It started by trying to understand whether AI understood Wayli.
Read the record →One of the hardest lessons wasn't deciding what to build. It was deciding what not to build yet.
Read the record →We weren't trying to build another product. We were trying to understand how AI would understand Wayli.
Read the record →The biggest breakthrough wasn't building another calculator. It was realising they all followed the same pattern.
Read the record →Building the Retirement Planner changed the question from 'When can I retire?' to 'When does work become optional?'
Read the record →While comparing mortgage overpayments with investing, we discovered that people weren't always trying to maximise wealth. Sometimes they were trying to buy time.
Read the record →Three simple scenarios revealed that the most important number wasn't the repayment. It was the point where the decision changed.
Read the record →I expected to spend most of my time writing code. Instead, I spent it learning how to ask better questions.
Read the record →The biggest improvement wasn't better code. It was asking better questions.
Read the record →AI helped me learn faster, but it never replaced the thinking behind the project.
Read the record →I thought I was building better calculators. I slowly realised I was trying to solve a different problem.
Read the record →BROWSE THE WORK
Topics
Things to Remember
If memory becomes infrastructure, Wayli does not need to remember everything - just what matters.
How do we remember what we decided if we let AI store our memory?
Proving that a system gives the same answer twice is not the same as proving that the answer is right.
A benchmark should be capable of persuading you to abandon the rule you were testing.
A USEFUL BOUNDARY
Labs records learning and evidence. It does not replace the shared method, Trust Centre, engine methodology or personal advice. Read each review according to what it actually assessed.