Blind Dev article

Lazy analysis: how not to waste 2–8 hours on a project

How I use AI to screen projects, verify mistakes and stop researching projects that do not interest me.

Archived post dated 2026-02-16. Service features, prices and market data describe the original publication period; their current status has not been verified.

Not investment advice. This article describes methodology and risks.

Translated from the original Telegram post, published on 2026-02-16. This is an archived account, not a current verification of services or markets.

Why I simplified project research

I used to spend several hours analysing projects. Part of that time went into writing articles: taking screenshots and explaining things takes time too. But even my Lighter livestream took about two hours.

Eventually I got tired of spending almost a whole day on one project, especially when dozens were appearing. I needed a simpler process.

I will not repeat the usual advice about visiting a project’s website, checking its quality or looking at the domain’s history and creation date to identify scams. Instead, here is how I work with AI.

My sequence

1. I send an AI model a long prompt based on the 4K framework. It must analyse the project’s team, concept, coin and code.
2. I receive a report and decide whether the project interests me at all. If not, I do not take the next step. This saves the most time.
3. I check for errors. When researching cyfer, for example, the model described the tokenomics of a different project with a similar ticker.

I also check the concept and documentation for completeness, quality and inconsistencies. If I find a mismatch, I ask again and investigate myself, or give the documentation to ChatGPT in Markdown for another check. Copying documentation takes about thirty minutes, but that is still faster than several hours of manual analysis.

Only if the inconsistencies are confirmed do I go to Discord and ask. If nobody answers, I skip the project.

AI usually finds the team information correctly, but the sources still need checking: are they about the right project, and is there any actual information? The tool you choose makes a substantial difference.

The original Telegram post included a file with the then-current version of my prompt. It changes as I find more precise ways to work with models, and as the market and my experience change. The prompt from that attachment is included at the end of this article.

Tools I used or planned to try

Surf

asksurf.ai finds team information well. It also studies the concept and documentation, though not without errors. As mentioned, it made a tokenomics mistake, but without similar tickers its prompt-based descriptions were generally good.

For code, I mean checking that repositories exist, that they are active, and that audits are available—not proving the code is safe. I was not sure how well it reviewed audit documents, so I downloaded the PDFs and uploaded them to ChatGPT for reading.

Surf can uncover team or project information I had not seen. That is useful when the linked sources confirm it and there is more than one supporting source.

Manus

Manus has an integrated browser. It studies documentation and SDKs and provides a Markdown summary for the requested topic. At the time of writing I had not used it for project analysis, but saw potential: on one occasion it searched for nine minutes and found the information I needed.

Parallel

I had heard that Parallel also performed research well, at a reported price of $0.30 per analysis. I had not tested it myself. This was a statement in the original post, not a verified current price.

There are enough tools to choose from. Better still, compare their answers.

When AI gets tokenomics or the concept wrong

For tokenomics, use Tokenomist, CryptoRank, the project’s documentation, or ask the team on Discord or Telegram.

For an accurate account of the concept, copy all the documentation pages into a Markdown file and give it to ChatGPT, requiring it to use only that file as its source. Another option is Perplexity, which also reports the number of sources it used.

P.S. I still write articles and posts almost entirely by hand; it is more reliable.

I hope you find good projects. What tools do you use for quick research? Do you use AI as a filter, or still analyse everything manually?

Prompt from the original post

Below is a complete English translation of the Russian attachment to the original Telegram post. This historical version uses cypher, cyphereth.com and token $CYPH as its example. For another project, replace the name, website address and ticker. The original criteria and scoring are preserved.

Evaluate the team, concept, coin and code of the cypher project: cyphereth.com (token $CYPH), assigning scores.
Each section starts with a default score of 1 out of 5. Each positive adds one point.
Use only verified sources and cite them.

Team: LinkedIn (information and professionalism). Present: +1 point; professional: another +1 point.
Social media (activity). Present: +1 point; is the information high-quality (details about the project and news, rather than meaningless content)? +1 point.

For the concept, examine the documentation: how unique and interesting is it (yes: +1)? Is there competitor analysis (yes: +1) and demand analysis (present: +1)? Are there technical details about the project (yes: +1 point) *?
* By technical details, I mean information about the consensus algorithm, software and node installation (for a blockchain), API (for any project), smart contracts (if it is not a blockchain), and so on. Do not forget to describe the main technical details according to what I have written here, if they are available.

Coin: tokenomics and initial allocation (how much goes to each category). If tokenomics is absent, the score remains 1, and only information about investments is considered. Ideally, the team should receive at most 10%, and investors too. More is bad; <= earns +1 point.
Tokenomics is available: +1 point; the distribution is decentralised (no more than 10% held by wallets, excluding staking, if such data is available): +1 point.
Unlocks are no more than 0.8% per month, with no sharp spikes (for example, 10% after a year): +1 point.
Are there investments, as well as well-known investors with a good reputation? +1 point.

Code: is it open? If yes, +1 point.
Is there a repository or contracts for the project's core functionality (software)? If yes, +1 point.
Is development active (yes: +1 point)? Are there audits (yes: +1 point)?