Authority graph // verified
Reputation, with receipts.
Not a logo wall. A public evidence ledger showing where Michał Piszczek has been quoted, where his ideas have been independently applied, and how original work travelled across markets and languages.
Original insight. External propagation.
The strongest reputation signal is not the first mention. It is what happens after an independent editor publishes the idea and other communities carry it further.
Named, quoted, attributable.
Newest first. These are the flagship records: an editor selected Michał as a named expert and published his argument under the outlet's own editorial responsibility.
AI could forever change the human psyche, experts say
George Lawton identifies Michał as CTO at Archdesk and cites his team's internal accounting that verifying AI-generated code can cost five to 30 times as much as generating it. The article also describes the team's practice of limiting code generation to what human reviewers can verify.
"People under pressure close that gap the easy way: they skip the check."
Who evaluates the AI evaluators?
Terri Coles uses Michał's argument across four sections of a reported feature on AI-led employee evaluation. Identified as Chief Technology Officer at Archdesk, he is cited on risk-tiered validation, verification budgets, worker challenges as ground truth and the blast radius of defective QA models.
"Tie the verification budget to the stakes of the decision, not to call volume."
Five expert contributions in twenty days
TechRound and its sister title Comms Link published Michał as a named expert five times in three weeks, each with headshot and attribution as Chief Technology Officer, Archdesk, across AI pricing, SaaS economics, security and infrastructure ownership. Contributions run verbatim, usually within a day of filing.
"AI didn't make attackers smarter. It made attacking cheaper."
"Community networks and local AI are the same economic story told twice."
Meta's new coding agent is cheap, but it will cost you your data
Three attributed observations on accepted-change cost, safe use cases and zero-data-retention boundaries for product code.
"The useful metric is cost per accepted change after review and repair."
AWS Introduces CloudFormation Express Mode for Faster Infrastructure Deployments
Michał is introduced as CTO at Archdesk and quoted on rollback risk in AI-assisted infrastructure.
"Rollback disabled by default is the detail people will skip past."
The ideas left the source.
These are stronger conceptual signals than reposts: independent practitioners apply the original work or cite an argument and extend it into a new operating context.
From a LinkedIn discussion to operating guidance
Across versions 1.2, 1.3 and 1.4 of its Observability module, the handbook names Michał Piszczek while turning his argument into operating guidance: validation evidence expires, ownership must survive personnel changes, and a resolved owner is not necessarily responsive.
The maintainer reports making CI checks required on main, including for administrators, after the discussion.
A Stolen Token Is Useless. A Stale One Still Works.
Kin Lane devotes a full response essay to Michał Piszczek's token-revocation framework. He applies its four planes, Revocation Completion Time and Revocation Exposure to Germany's federation blueprint and to his own API Authorization Profile.
Independent analysis, named attribution and two canonical backlinks. The framework is used to audit a real architecture, not merely mentioned.
Measuring energy per accepted task
Xiong, whose profile lists a Principal Engineer role at Intel, quotes Michał Piszczek's method for measuring p95 latency, completed jobs per hour at an agreed quality threshold and whole-server energy. Xiong extends it with a definition of the job and describes building a tool around the method.
"So this is the method I'm building my next tool around:"
Declared method reuse; a working implementation and independent measurements have not been verified.
The facts layer shouldn't be a mirror
Fulnecky names Michał Piszczek, quotes his response about generated diffs and postmortem ownership, then extends it into a case for recoverable reasoning trails beside executable engineering rules.
"Git blame still means something even when the diff was generated."
66%
Multiplying four 90% proof gates leaves roughly 66% accepted autonomy
The author names Michal Piszczek, Archdesk CTO, explains Verification Cost and the complete Proof-Adjusted Autonomy formula, then applies the framework to automation and quality control in Japanese e-commerce operations.
One independent essay, published on two platforms. Counted as one adoption event, not two.
Owned work, independently selected.
Newest first. These are posts and essays published by Michał, then selected by a platform's editorial team. They are strong distribution signals, but they are not counted as independent quotes or interviews.
What Is RSI AI? Recursive Self-Improvement Explained
Picked up by a HackerNoon editor on publication day. HackerNoon records the original piszczek.pl essay on recursive self-improvement as its first-seen source.
When AI Makes Skill Abundant, Scarcity Moves Somewhere Else
HackerNoon's editors selected the syndicated essay as a Top Story. The article names Michał as author and records the original piszczek.pl essay as its first-seen source.
How I Got Qwen3.8 27B Running at 50 Tok/s With 256K Context
The Qwen3.8 field report was picked up by a HackerNoon editor and published the same day. The article links back to the interactive piszczek.pl demos it benchmarks.
Muse Glimmer, China's open-weight lead and useful intelligence per watt
LinkedIn News included Michał's Muse Glimmer post in its editorial storyline. The selected post connects open-weight competition, 24 GB local inference and Joules per Verified Task.
Research tested and referenced.
Tool makers, videos and commercial blogs that test or reference Michał's published work, credited by name or by his cdiamond release. Tests, recommendations and secondary commentary are labeled separately from direct editorial features.
Qwen 3.8-27B on Cheap Hardware - Can It Handle Full 262K Context?
Uses Michał's ten-run Qwen3.8 benchmark and his findings on drafter fit and testing an occupied context. The video distinguishes the 50.44 tok/s production average from 12.61 tok/s at a filled 256K cache. Its description names him and links to his HackerNoon report.
A small specialist channel discussing published results. This is secondary coverage, not an independent rerun of the benchmark.
All AI Benchmark also uses the 12.61 tok/s result at 261,500 tokens to compare published hardware tests. The configurations differ, so the comparison is directional.
Open-source benchmark contribution. In llama.cpp PR #25635, the PR author thanked Michał for testing on an RTX PRO 4000 Blackwell and cited his 32K results: +7.4% prefill throughput and +2.6% hot decode throughput.
Qwen3.8 GGUF: tested by other tool makers
Michał Piszczek's GGUF quantization, published as cdiamond, has been tested by three tool makers. The base model is Qwen's; Michał developed this mixed-precision build and its benchmark configuration.
Atomic Chat recommends the exact file in its RTX 5090 guide after comparing it with its own AD-Q4 build. llmash reports single-stream results on an RTX PRO 6000 Blackwell. MASHBOX lists a short Quick check on TPU v6e-1.
Technical references. Aexol AI restates the published throughput figures. OpenClaw DC cites the model card's perplexity controls.
Public AI analysis, cited in product research
Three articles attribute Michał's LinkedIn analysis and a comment: Argon vs. Sonnet at the same price, the selection of vendor benchmarks, and whether a HubSpot renewal outlasts an OpenAI spending commitment. Both Argon articles draw on the same original post.
eesel says it uses its own AI writer for its blog. These are attributed citations in commercial content, not interviews or editorial awards.
Propagation without self-publishing.
Newest first where a publication date is verifiable. These records repeat, translate or syndicate an editorial quote or an original benchmark, but are not presented as new independent interviews.
Not everything is a press citation.
Vendor case studies, historical reporting and news aggregation add context to the public record.
DeepInfer discussion alongside Bloomberg coverage
Techmeme included Michał Piszczek's LinkedIn analysis of DeepInfer in its roundup of reactions alongside Saritha Rai's Bloomberg article.
How Archdesk redesigned a complex Construction Tech SaaS application
A vendor portfolio case study containing a named testimonial from Michał Piszczek, CTO of Archdesk. Useful product-leadership evidence, not independent editorial coverage.
Sun Microsystem pouczony przez Polakow
An archived report documenting the responsible disclosure of critical issues affecting Sun Microsystems. It anchors the public technical record more than two decades before the current AI work.
The count means something.
A reputation page loses value the moment it treats every scrape, mirror and owned profile as an independent endorsement.
Included and labeled
- Named editorial quotes and interviews
- Platform editorial picks of owned work
- Independent applications of a published concept
- Practitioner citations that extend the original argument
- Translations and syndication with clear source lineage
- Case studies and historical records with context
Not inflated into coverage
- Owned cross-posts and personal profiles
- Articles with a similar thesis but no attribution
- Pitches that did not produce a public citation
- Multiple mirrors counted as separate adoption events
Need a source, quote or technical argument?
Use the media kit for bios and booking, or the citation center for DOI-backed concepts, software and formulas.