# thekb.eu — Tech watch on AI, coding agents & the SDLC

> Knowledge base of analytical "fiches de veille": 401 fiches, knowledge graph of 3304 entities / 6302 triples. Default language: English (source: French). Every fiche page has a Markdown twin at {url}.md.

> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Free to quote, reuse and train on — attribution to theKB.eu (with a link) is required.

## Recent fiches
- [Claude Fable 5.1 and Mythos 5.1: pricing, safeguards, benchmarks](https://www.thekb.eu/en/fiches/anthropic-claude-fable-5-1-mythos-5-1-2026-09-01.md): Product communication from **Anthropic** published on **September 1, 2026** on anthropic.com (~4,000 words, six sections, 22 testimonials from early-access partners). It announces **Claude Fable 5.1** (general availability) and **Claude Mythos 5.1** (verified access): *the same model, but with different levels of safeguards*.
- [AI Agent Agency: The Hugging Face Incident and Twilight Factory](https://www.thekb.eu/en/fiches/mollick-agency-and-agents-twilight-factory-2026-08-31.md): Post by **Ethan Mollick** published on **August 31, 2026** on *One Useful Thing* (~2,200 words). He starts from a security incident to raise an organizational question: when should an AI ask a human for help?
- [AI Equity, Job Loss, and the Case for Human Reserved](https://www.thekb.eu/en/fiches/gates-ere-ia-turbulente-choix-critiques-2026-08-26.md): Essay published on **Gates Notes** on **August 26, 2026** by **Bill Gates**, co-founder of **Microsoft** and chairman of the **Gates Foundation**, ~4,500 words, announced as the first in a series. The text poses an alternative — AI will be the greatest equalizer ever invented, or the worst source of injustice — and a finding: no plan exists for entering this period. **(A) Three risks**: the lasting disappearance of entry- and mid-career jobs, white-collar as much as blue-collar, within a decade rather than several generations, because this time the substitution targets **cognition**; the weaponization of malicious actors (cyberattacks, bioterrorism, fraud, deepfakes), coupled with a concentration of power among those who already hold it; the effect of compagnons IA on children's development and on critical thinking. **(B) The benefits**, located in five domains — research, health, agriculture in low-income countries (the impact the author calls the fastest), public services, education — with a reservation carried on the verb: *"the operative word is 'can'"*. **(C) Three proposals** open the series: building an unprecedented national and international institutional framework, borrowing from the nuclear inspection regime, aviation regulation, and ozone agreements; reserving certain occupations for humans, a domain named **Human Reserved**; **taxing AI tokens and robots** to rebalance the taxation of labor and capital. Gates discloses his financial ties to the industry and the transfer of his profits to the foundation. The text extends executive essays on the distribution of AI's value — [[nadella-frontier-ecosystem-human-token-capital-2026-06-12]], [[zuckerberg-meta-future-is-for-everyone-superintelligence-2026-08-10]] — by focusing on public power rather than the firm.
- [DuckDB, DuckLabs, and AWS: the Changing Physics of Analytics](https://www.thekb.eu/en/fiches/warfield-duckdb-changing-physics-analytics-2026-08-26.md): Guest post by **Andy Warfield**, an engineer on the **S3** team at **AWS**, published on **August 26, 2026** on *All Things Distributed*, **Werner Vogels**'s blog, who introduces it in a few lines signed "--W": **3,554 words** per the page. The text serves as the vehicle for the announcement that **DuckLabs**, the team behind **DuckDB**, is joining **AWS**. (A) The thesis: systems computing is about seeking the elegant trade-off against a moving "physics" — the ratios between memory speed, network, and compute — and that physics has changed. Warfield quantifies the gap: an **m1.xlarge** from 2007 offered **15 GB of RAM**, **4 virtual cores**, and **~1 Gb/s** of network; an **m8g.48xlarge** today offers roughly **50×** more of each of the three. Dataset growth, meanwhile, follows a distribution whose tail consists of very large volumes. (B) The consequence: distributed processing — **MapReduce**, **Spark**'s **RDDs** — was designed under the I/O constraints of the early 2000s, and much of the work assigned to it no longer needs to leave the application. Hence the embedded, in-process library engine, running in the application's address space, of which **DuckDB** is the example. Warfield anchors this in the *Scalability! But at what COST?* paper (2015) and **Paul Barham**'s epigraph: "You can have a second computer once you've shown you know how to use the first one." He states an explicit caveat: "When a job genuinely needs a thousand machines, it needs a thousand machines." The corpus already holds [[vogels-tech-predictions-2026-allthingsdistributed-2025-11-25]] from the same blog and [[anthropic-self-service-data-analytics-claude-agentic-stack-2026-06-03]] on self-service analytics.
- [When Code Is Abundant, Trust Is the New Bottleneck](https://www.thekb.eu/en/fiches/staples-gitlab-when-code-is-abundant-2026-08-24.md): Essay by **Bill Staples**, CEO of **GitLab**, published on **August 24, 2026** on the about.gitlab.com blog: an announced **31-minute** read, roughly **39,000 characters**, presented as the sequel to a memo written to the board of directors in January 2026 and partially published in May under the title *GitLab Act 2*. The text presents itself as a response to **Anthropic**'s AI-native SDLC playbook, published three days earlier, from which it borrows the opening line — "Code is no longer the bottleneck" — to pose the question that drives it: what becomes scarce when code becomes abundant. (A) The economic diagnosis: the useful unit is not cost per line but **cost per accepted change**, which aggregates generation, environment, context, verification, review, remediation, and governance; AI collapses the generation term alone, which makes the others proportionally heavier — an organization ten times faster at generating "will simply move the queue." (B) The architectural response: four capabilities — agent platform, machine-scale execution, durable context, governance — forming an enterprise layer that outlives the model, "The model should be replaceable. The agent should belong to the customer." (1) Three modes coexist durably, from human-driven legacy to autonomous development, against the idea of a single maturity curve. (2) The CI/CD pipeline becomes the place where the inner loop runs, instead of being an end-of-line gate. The figures cited are Stripe's, Spotify's, and Amplitude's; GitLab produces only one, about its own source control. The corpus already holds [[claxton-anthropic-ai-native-sdlc-playbook-2026-08-21]], the source this text responds to, and [[sfeir-sdlc-pdlc-articulation-2026-07-22]] on the SDLC/PDLC articulation that Staples adopts as his own.
- [AI-Native SDLC: From intent.md to bands.yaml Governance](https://www.thekb.eu/en/fiches/claxton-anthropic-ai-native-sdlc-playbook-2026-08-21.md): Long-form guide from **Anthropic** by **Louis Claxton** (Applied AI team), published on **August 21, 2026** on the claude.com blog: a stated **40-minute** read, roughly **64,000 characters**, presented as a collection of *plays* drawn from the team's work with its clients. (A) The diagnosis: with code no longer the bottleneck, it shifts to the stages on either side of the build (plan, review/test, deploy), line-by-line controls stop holding once the agent writes most of the diff, and governance cost rises as exceptions still route through periodic committees. (B) The response: six stages (Plan, Design, Build, Test, Deploy, Maintain) organized as a **loop** rather than a chain, each ending with a **committed artifact** that the next stage reads — `intent.md`, `spec.md`, `plan.md`, the diff and its tests, the PR and its findings, the incident record. (1) Institutional knowledge becomes versioned files: `CLAUDE.md`, skills, `REVIEW.md`, `bands.yaml`. (2) Governance splits into two layers, with the skill positioned as an advisory control and the hook as the deterministic layer behind it. Separation of duties is set as an invariant — the agent that writes the code cannot approve it — and the piece closes on *"The loop keeps running. Human judgement stays above it."* The corpus already holds [[clinton-anthropic-secure-ai-native-sdlc-2026-07-21]] on the security side of the same cycle, and [[hingel-augment-how-ai-changes-sdlc-six-stages-2026-06-08]] on the same six-stage breakdown as seen by a competitor.
- [How Startups Ship With Claude Code: Five Rules](https://www.thekb.eu/en/fiches/segner-anthropic-claude-code-guide-startups-2026-08-20.md): Guide signed by **Michael Segner**, published on **August 20, 2026** on the claude.com blog in the *Claude Code* category: a **5-minute** read announced for approximately **31,500 characters** of body text, also offered as a PDF. Stated material: interviews with **more than a dozen** startups, fifteen named — **Artemis Security**, **Cainex**, **Clay**, **ClickHouse**, **Cognition**, **Commure**, **Crosby**, **Emergent**, **Harvey**, **Heidi**, **Higgsfield**, **Omni**, **Parahelp**, **Translucent**, **Zingage**. (A) Five operating rules: *everyone ships*, *automate the tedium*, *trust, but verify*, *build for rebuilding*, *prototype, dogfood, productionize*, each closed with product tips and gathered into a final checklist. (B) A body made of attributed quotes, each rule illustrated by named executives rather than by an aggregated metric. The four figures highlighted are those of the interviewed companies: **+30%** more features shipped (ClickHouse), **2 to 3×** engineering productivity (Omni), **100%** of bug triage automated (Clay), **more than 6,000 PRs per week** (Artemis Security). Two passages depart from the testimonial register: **Cainex**'s self-correction loop on medical coding, described step by step, and the internal use of **Claude Tag** at **Anthropic** as first responder for CI/CD on-call. The question posed at the opening — *"what would it look like if an organization built their product development lifecycle with Claude Code from the ground up?"* — connects with [[claxton-anthropic-ai-native-sdlc-playbook-2026-08-21]], published the next day by the same publisher, and extends [[cherny-wu-reflecting-year-claude-code-2026-07-17]].
- [Block Open-Sources Berd, Its Agent Desktop Client](https://www.thekb.eu/en/fiches/block-berd-caractere-agents-open-source-2026-08-18.md): Corporate blog post from **Block** (`block.xyz/inside`), unsigned — the displayed author is **"Block"** —, published on **August 18, 2026**, ~930 words, announcing **the open-sourcing of Berd**, Block's internal desktop application for working with agents, and laying out the design thesis that guided it: giving agents character *"not only through roles, instructions, skills, and tools, but through distinctive visual identities"* — hence the in-house animated characters, the *"Gloopies"*. The post starts from an observation of fragmentation (*"The technology was powerful, but the experience around it was fragmented"*) and a precisely named interface problem: *"the product gives people little sense of how the agent is configured, which context and tools are available to it, and how it differs from another agent"*. Two structuring contributions. **(A) A three-tier articulation**: **goose** remains the framework and *runtime* that holds the agent loop; **Berd** is the desktop client (projects, context, sessions, agents, configuration); the two communicate via the **Agent Client Protocol**. **Buzz** is designated as the follow-up, for when solo work becomes collaborative (*"Start alone, then go multiplayer"*). **(B) Six requirements handed off to Buzz**, stated as a takeaway: *"private space, durable context, recognizable agent identities, reusable skills, visible configuration, and clearer visibility into an agent's configured context, tools, and capabilities"* — a grid directly reusable for evaluating an agent client. The text itself distinguishes identity from capability: *"The avatars make the agent recognizable. Its role, skills, and tools make it useful."* No usage figures are produced and no license is named for the open-sourcing.
- [AI-Era Software Supply Chain Risk: Sonatype's 49-Month Study](https://www.thekb.eu/en/fiches/linskens-sonatype-securite-vitesse-ia-quatre-ans-2026-08-18.md): Blog post from **Sonatype** by **Aaron Linskens** (*technical writer*), published on **August 18, 2026**, ~1,300 words: it recounts a **Sonatype Research Labs** study spanning **49 months** (June 2022 — June 2026) and a **fixed cohort** of enterprise applications, a methodological choice asserted to isolate the evolution of the application fleet rather than that of the customer portfolio. The result is presented as a contradiction: remediation is faster, yet risk accumulates further. (A) **The stock is rising** — *Critical* and *High* vulnerabilities per application **×4.31** (from **14.14** in June 2022 to **54.3** in 2026, still **×3.91** excluding legacy applications newly brought under management), newly affected component versions at **46×** the pre-AI rate, monthly application creation **×4.84**. (B) **Remediation is improving** — more than half of resolved violations are resolved in under a day, the median age of unresolved *Critical/High* vulnerabilities drops from **228** to **126 days**, then to **103** in May 2026; among cohorts that had twelve months, **52.6%** are resolved, **44.3%** open, **3.1%** under waiver. (C) **The proposed lever is component selection**: at the moment a vulnerable dependency was chosen, a substantially less risky version already existed in **62.2%** of cases on **Maven**, **46.9%** on **npm**, **34.3%** on **PyPI** — a gap the text attributes to an information gap rather than developer fault. The post itself states that AI is not the sole cause of the acceleration, and concludes on **Sonatype Guide**, which brings this intelligence to the point of selection. On the supply-chain side, it extends what [[fiches/2026-08/staples-gitlab-when-code-is-abundant-2026-08-24]] frames in economic terms and [[fiches/2026-07/clinton-anthropic-secure-ai-native-sdlc-2026-07-21]] in secure-cycle terms.
- [Buzz Projects: Git Forge on Your Own Nostr Relay](https://www.thekb.eu/en/fiches/petersen-block-buzz-projects-forge-souveraine-2026-08-18.md): Product announcement post from **Block Engineering** signed by **Thomas Petersen** (*Principal Designer & Builder*), published on **August 18, 2026**, ~1,800 words across thirteen short sections, introducing **Buzz Projects** — a **software forge hosted on its own relay**: Git repositories, branches, pull requests, issues, review and merge, multi-repo projects, an activity feed, all linked to conversation channels. The post's standfirst and thesis: *« Coding agents are the terminal for your computer. Buzz is the terminal for your network. »* Three contributions. **(A) A trust doctrine grounded in *ex post* proof rather than *ex ante* authorization**: on one side *« No forced guardrails, no limitations on what your agents are allowed to help you with »*, on the other *« Every push, review, approval, and merge is a signed Nostr event. If an agent authors a patch, you can see which agent produced it and which human authorized that agent to act »*; the section closes on a stated direction — *« we are already exploring ideas around agent trust protocols informed by past behavior »*. **(B) Git interoperability without proprietary tooling**: *« These are standard git repositories… You can fetch, clone, pull, and push over plain Smart HTTP, with no custom tooling or wrapper CLI required »*, with the clé Nostr serving as a single identity — *« The same npub that signs your messages signs your pushes. »* **(C) A distinction between execution surface and network presence**: *« A terminal gives an agent somewhere to execute commands and change files, but it does not give it a persistent place in the network. Buzz does. »* The post produces no figures and contains no outbound links; it qualifies itself as preliminary six times (*« still very basic »*, *« fairly elementary »*, *« still under experiments »*), and Projects lives under the **Experiments** tab of Buzz Desktop.
- [Andrew Ng's AI Engineering Skills Map: Reskilling, Not Roles](https://www.thekb.eu/en/fiches/ng-ai-engineering-skills-map-2026-08-14.md): X post by **Andrew Ng** from **August 14, 2026** (16:29 UTC), reprising the "Dear friends" letter from ***The Batch* #366** (DeepLearning.AI, same date), ~900 words. Ng presents **The AI Engineering Skills Map** and publishes **four skills** held to be the most important. **(1) Building and deploying AI applications** — the specificity is named: *« The key difference between AI and non-AI applications is that the former has unpredictable outputs »*, hence the emphasis on *evals* and error-analysis loops. **(2) Software engineering fundamentals**, because *« Understanding software fundamentals allows you to recognize what tradeoffs even exist »* — the inexperienced developer fails *« because they don't know what context to give their coding agent »*, hence the goal of *« steering coding agents using the precise language of software engineering »*. **(3) Using coding agents**, in an operational formulation: *« help the agent autonomously close loops by providing verifiers or evals »*, and *« knowing how much to intervene and how much to leave them alone »*. **(4) *Shaping the build***: *« Given a clear spec, coding agents are rapidly improving at delivering to it. Thus, our work as engineers is shifting toward deciding what should be in the spec »*, paired with *« Engineers should no longer expect to be given a pixel-perfect design and asked only to implement it. »* A **terminology note** carries most of the framing: Ng talks about **skills** in AI engineering and **not the role** "AI Engineer", with an explicit analogy — *« All developers today should know how to work with the cloud, and only a smaller number have a "Cloud engineer" title. »* The whole is backed by *« an analysis of more than 10,000 job postings, dozens of structured interviews with experts, hiring managers, and recruiters, surveys, and other online data »*, of which **no numeric results are published**: Ng describes his process as *« informally… akin to running clustering »* and announces a detailed map in future posts. He states the interest in the second-to-last sentence: *« DeepLearning.AI's principal focus is to help developers gain these AI engineering skills. »*
- [GLM-5.3: Post-Training Scaling Yields Cyber Capability](https://www.thekb.eu/en/fiches/zai-glm-53-emergent-cyber-2026-08-14.md): Announcement post published on the **official Z.ai blog** (formerly Zhipu AI, Chinese lab) on **August 14, 2026**, **with no individual byline**, ~2,000 words plus footnotes. It announces **GLM-5.3**, successor to GLM-5.2, opening with a methodological thesis: *« Scaling post-training is all we did for GLM-5.3. »* Same base model as GLM-5.2 — *« every gain comes from post-training »*. Three announcements. **(A) An open-weights coding model**: +50% claimed on **Z.ai Code Bench**, an unpublished in-house benchmark. **(B) A cyber capability presented as "emergent"**, which the body of the text traces to a training choice — *« As part of post-training, we introduced vulnerability discovery data and environments into the training mix. We expected this to make the model better at finding and reasoning about vulnerabilities »* — what came as a surprise was the speed and the change in nature: the model moves from identifying isolated flaws to *« coherent plans for complete exploitation chains »*. Gains grow with position in the exploitation chain: CyberGym 77.2 → **84.5%**, ExploitBench 24.4 → **54.4%** (×2.2), ExploitGym 29 → **105** tasks in 2h (×3.6), with the gap to the closed frontier remaining wide (181 and 247 tasks). Z.ai puts it this way: *« Capability is growing fastest exactly where we are furthest behind. »* The post also publishes a **Z.ai Security Disclosure Ledger**: **2,436 vulnerabilities identified across 269 open source projects** — kernels, OSes, browser engines, infrastructure, web applications, network protocols — the oldest introduced in **1981**, average lifetime before discovery **26.6 years**, of which **53 disclosed** and **2,383 under embargo**. **(C) A weight release** *« within two weeks of launch, once safety evaluation and hardening are complete »*. The most reusable methodological contribution: **environment and verifier synthesis**, the latter produced without access to the reference solution and admitted only after a triptych of negative controls — **oracle**, **no-op**, **unsolved-state**. All agentic evaluations are conducted **in Claude Code 2.1.207**.
- [DeepSeek Harness (dsh) developer preview explained](https://www.thekb.eu/en/fiches/deepseek-harness-everything-is-a-plugin-2026-08-13.md): Official product page from **DeepSeek**, published on **August 13, 2026**, **unsigned**, ~450 words, announcing the *developer preview* release of **DeepSeek Harness** (`dsh`) — a coding-agent harness **open source under the MIT license**, whose repository opened the same day. A three-word thesis, repeated in the title and in the repository description: *« Everything is a plugin »*, paired with a second promise, *« Every run is traceable »*. The page states the equation *« AGENT = MODEL + HARNESS »* and lists the pluggable capabilities — *« models, tools, skills, sessions, sandboxes, storage, loops, scheduling, and the UI »*. Four modes ship: **Standard** (full coding agent), **Code** (tools exposed via the *Code Mode SDK*, letting the model compose multi-step operations inside a TypeScript program), **Minimal** (*« two-tool coding agent with persistent bash and str_replace_editor »*, explicitly *« for benchmarking models in a minimal environment »*), and **Creator** (runtime inspection, in-memory plugin testing). The technical substance sits in the repository, not on the page: `docs/architecture.md` states a logging invariant — *« Model-visible means logged. Anything that reaches a model request must be reconstructable from the log, and a runtime invariant asserts it »* — and states that *« there is no privileged core to patch »*. The technical core is not DeepSeek's own: DSH is built on **Cordis** (the `cordiverse` project, a third party), **vendored** into `vendor/` with a manifest and a sync procedure, and the page places the *« Cordis paper »* at the same navigation level as "GitHub" and "Developer docs". Two LLM adapters ship — `dsh-llm-deepseek` and `dsh-llm-pi-ai`, a generic multi-provider adapter. The repository warns in capitals: *« THERE WILL BE COMPATIBILITY-BREAKING CHANGES »*, and `CLAUDE.md` specifies that `SESSION_FORMAT_VERSION` stays at `0` *« with no compatibility promise »*, with backends rejecting old on-disk formats. Timeline: DSH ships on the day **DeepSeek-V4-Pro reaches GA**, three days before a new API pricing schedule takes effect on **August 16, 2026 at 16:00 UTC**, with peak/off-peak rates and an off-peak discount of **−50%**.
- [Buzz (buzz.xyz): Block's Nostr Agent Workspace, Explained](https://www.thekb.eu/en/fiches/buzz-block-panorama-deep-research-2026-08-12.md): Internal research report dated **August 12, 2026** consolidating, for presentation purposes, everything publicly documented about **Buzz** — **Block**'s humans + agents workspace, launched on **July 21, 2026** under the **Apache 2.0** license. It aggregates the two engineering posts already filed alongside the corporate announcement, the GitHub repository, press coverage, X, and **three independent hands-on accounts** that constitute the dossier's only non-self-reported data. **(A) A vocabulary gap documented by quotation**: **Jack Dorsey**'s launch tweet announces *"model-agnostic, decentralized, self-sovereign, and open source"*; Block's `ARCHITECTURE.md` states *"The relay is the single source of truth. All reads and writes flow through it. There is no peer-to-peer event exchange, no gossip, no replication."* The relay is therefore single and authoritative per community: Buzz's "decentralization" is an **organizational sovereignty** — self-hosting and portable identity — not network redundancy. **TFTC**'s formulation: *"Two of those three hold cleanly. The third needs a qualifier."* **(B) An asymmetry between demonstrated rigor and exploitation risk.** On one side, a rare degree of formalism for a v0.4.x/0.5.x: multi-tenant isolation specification **mechanized in TLA+**, authorization properties verified in **Tamarin**, a model-checked Git storage protocol, a hash-chained append-only audit log, 127 *event kinds*, NIP-01/42/98/34. On the other, channel membership is the unit of permission — *"channel membership is not fine-grained tool authorization"* (João Queirós) —, agents run in `--dangerously-skip-permissions` outside any sandbox on a human's machine, and observability is lacking: *"Buzz tells me an agent got a message. It doesn't tell me what happens next"* (DevTools Daily, which reports silent OOM kills). Block acknowledges it: *"the agent can do anything, and security rests entirely on restricting who can tell it what to do"*. **(C) The technical stack**, absent from the filed posts: **Rust** relay (Axum WS + REST), **Postgres**, **Redis**, **S3/MinIO** via Blossom, **Tauri + React** desktop client. Agent integration goes through **`buzz-acp`**, an **ACP** harness that plugs in goose, Codex and Claude Code and translates **ACP ↔ MCP**, plus **`buzz-agent`**, an in-house agent. The report corrects itself on one point: the *"+33% more work"* in Block's TL;DR is the **ratio of completed tasks (20 versus 15 out of 44)**, not a score gain — the score itself rises from 59.1% to 71.5%, i.e. **+12.4 points**.
- [ChatGPT Desktop & Claude Desktop vs Web: Fact-Checked](https://www.thekb.eu/en/fiches/chatgpt-claude-desktop-vs-web-deep-research-2026-08-12.md): Internal research report dated **August 12, 2026** (in *What? — So What? — Now What?* format, investigation conducted August 11-12) on a simple question: are the **desktop** applications of ChatGPT and Claude better than their **web** versions? The answer comes in two parts. **(A) A solid, well-sourced qualitative consensus exists.** The starting point is indisputable: desktop and web call exactly the same cloud models, the application being merely an interface to the service — the gain therefore lies entirely in the application shell (access latency, stability during long sessions, memory footprint, system integrations, workflow fluidity). What genuinely distinguishes desktop, confirmed: on the OpenAI side, a global shortcut (Option/Alt + Space), a *companion window* that always stays on top, native screenshots, and since July 2026 the **Codex/Work** agentic capability built into the app; on the Anthropic side, **Quick Entry** (macOS), **Desktop Extensions** (installing a local **MCP** server becomes *"as simple as clicking a button"*), access to local files, **Cowork** and **Computer Use** (Accessibility permissions and screen recording). The web retains two confirmed strengths: multiple tabs/threads, and universality without a client to install. **(B) Nearly all the figures circulating to support this consensus do not withstand verification.** The report's critical audit (§1.5) classifies **unconfirmed** seven widely repeated numerical claims: the *cold start* "2-3 s vs 8-12 s" (the only trace being an anecdotal *"loads in about 3 seconds"* on Substack); RAM usage "200-700 MB vs 1.2-2 GB," attributed to an "Alibaba Product Insights" whose pages return **404**; an untraceable glitch rate and session retention figure; a "Claude +10-20% end-to-end" attributed to **Skywork**, which had in fact benchmarked its own Windows agent rather than Claude against the web; an untraceable "Cosmo Edge" source; unconfirmed Zenken AI citations; and two unauthenticated X posts with no URL. The counter-signal is documented with the same rigor: Yuri Dvoinos describes a Claude Desktop app that *"makes me want to throw my laptop out the window"* — 68% CPU usage, input lag on a MacBook Pro — and the report notes that both apps are **Electron** builds with native layers. Hence its formulation: *the desktop advantage is a promise of implementation, not a law of nature.* **The "So What"**: since the model has become the common denominator, the interface becomes the battleground — the **Codex + ChatGPT** merger of July 9, 2026 and the Cowork/Computer Use tandem tell the same story, *"the desktop app is no longer a chat client, it's an agent runtime with access to the machine."* Three consequences: the gain is a **friction** gain, not a power gain; for a CIO, desktop **shifts the trust boundary** — Computer Use requires sensitive system permissions and the Codex merger places code execution, browser, and connectors within *"one expanded trust boundary,"* whereas the browser remains governable via SSO, DLP, and CASB; and for anyone publishing, the fragility of the figures is itself the story. **The "Now What"** delivers individual switching criteria, a CIO checklist (inventory permissions, disable Computer Use and Cowork by default, scope which MCP extensions are authorized, organize distribution and updates — on Linux, outside the apt repository, Claude Desktop does not update itself) and an editorial directive: cite only confirmed verbatims and dates.

## Topics
- [AI Coding Agents & Skills](https://www.thekb.eu/en/topics/coding-agents-skills/): Coding agents, skills, prompting craft and agent-native workflows.
- [Architecture & Construction](https://www.thekb.eu/en/topics/architecture-construction/): Software architecture and how systems get built in the AI era.
- [Transformation & Adoption](https://www.thekb.eu/en/topics/transformation-adoption/): How teams and organizations adopt AI-assisted development.
- [Quality & Security](https://www.thekb.eu/en/topics/quality-security/): Testing, review, reliability and security of AI-produced software.
- [Economy & Market](https://www.thekb.eu/en/topics/economy-market/): Economics, pricing, market moves and the business of AI tooling.
- [Philosophy & Society](https://www.thekb.eu/en/topics/philosophy-society/): Philosophy, cognition and societal effects of AI.
- [Strategy & Frameworks](https://www.thekb.eu/en/topics/strategy-frameworks/): Strategic frameworks and doctrines for the AI transition.
- [Tools & Platforms](https://www.thekb.eu/en/topics/tools-platforms/): Tools, platforms and infrastructure of the agent ecosystem.
- [Research & Education](https://www.thekb.eu/en/topics/research-education/): Research findings and how we learn to work with AI.
- [Products & Services](https://www.thekb.eu/en/topics/products-services/): Products and services reshaped by AI capabilities.
- [Policy & Regulation](https://www.thekb.eu/en/topics/policy-regulation/): Policy, regulation and governance of AI systems.

## Glossary (key concepts)
- [ACE](https://www.thekb.eu/en/glossary/ace/): ACE — an agent context-engineering method reported to improve agent accuracy by about 10.6% and finance-task accuracy by 8.6% while cutting latency by roughly 87%. It illustrates how structuring the context supplied to an agent, rather than changing the model, can lift measured performance.
- [ADLC](https://www.thekb.eu/en/glossary/adlc/): Agentic Development Lifecycle — a proposed software lifecycle designed around the properties and failure modes of AI models rather than human teams. Each phase maps to a defended failure mode or an exploited model property, structured as phases and verification gates. Cited as a convergent framework reached independently by several practitioners.
- [AEO](https://www.thekb.eu/en/glossary/aeo/): Answer Engine Optimization — the practice of structuring content so AI answer engines cite it when responding to queries. It adapts search-optimization thinking to a world where users read a generated answer rather than a list of links, prioritizing citable, well-attributed statements over keyword ranking alone.
- [Agentique adaptative](https://www.thekb.eu/en/glossary/agentique-adaptative/): An architectural approach for putting agentic AI into production, organized around four pillars. It addresses how autonomous agents are structured, supervised, and adapted so they keep working reliably under real operating conditions — with edge cases, drift, and load — rather than only in controlled demonstrations.
- [agents parallèles](https://www.thekb.eu/en/glossary/agents-paralleles/): A working pattern in which many agent instances run at once on the same effort — reported at up to 16 simultaneous agents across roughly 2,000 coding sessions. Running agents in parallel raises throughput and lets independent sub-tasks proceed together, at the cost of coordination and oversight.
- [AGI](https://www.thekb.eu/en/glossary/agi/): Artificial General Intelligence — a system able to match or exceed human capability across most cognitive work rather than one narrow task. Framed by leading AI companies as their stated goal, often defined by benchmarks such as PhD-level research or self-improvement. Whether current approaches can reach it is contested.
- [AI-Assisted Engineering](https://www.thekb.eu/en/glossary/ai-assisted-engineering/): The methodical integration of AI into a mature software development lifecycle, aimed at code that stays secure, scalable, and maintainable. It positions AI as one disciplined stage within established engineering practice rather than a replacement for review, testing, and design.
- [AI brain fry](https://www.thekb.eu/en/glossary/ai-brain-fry/): Mental fatigue from excessive use or oversight of AI tools beyond one's cognitive capacity. Reported symptoms include a buzzing sensation, mental fog, and slower decision-making. The term names a human cost of continuous agent supervision, distinct from the productivity gains usually emphasized.
- [AI slop](https://www.thekb.eu/en/glossary/ai-slop/): Low-effort AI-generated output produced without real understanding or review — in software, code that compiles and may pass tests yet degrades a codebase's clarity and long-term quality. The term is pejorative, marking the gap between genuine engineering and volume generation, and warns against accepting agent output uncritically.
- [AI4Ops](https://www.thekb.eu/en/glossary/ai4ops/): The application of AI to IT operations, oriented toward autonomous operation of infrastructure and services. It extends the automation of monitoring, incident response, and remediation, so that operational tasks are increasingly handled by agents rather than triggered manually by an on-call engineer.
- [approche spec-driven IA](https://www.thekb.eu/en/glossary/approche-spec-driven-ia/): A spec-driven method for AI-assisted development structured in stages: onboarding, atomic planning, iterative development, and capitalization. Work is anchored on an explicit specification the agent follows, so intent is fixed up front and progress accumulates into reusable knowledge rather than one-off output.
- [augmented coding](https://www.thekb.eu/en/glossary/augmented-coding/): AI-assisted coding that keeps quality, testing, and coverage as first-order priorities. It contrasts with looser styles that accept generated output uncritically: the developer stays responsible for correctness, using the agent to move faster without ever lowering the bar on verification and review.
- [Augmented Craftsman](https://www.thekb.eu/en/glossary/augmented-craftsman/): A developer augmented by AI who nonetheless stays in the code — reviewing, shaping, and owning the result rather than delegating it wholesale. The term marks a stance that keeps human craft and judgment central even as agents handle more of the mechanical work.
- [BMAD](https://www.thekb.eu/en/glossary/bmad/): BMAD (Breakthrough Method for Agile AI-Driven Development) — an agile methodology for AI-assisted software work, described through the metaphor of an urban plan for agentic AI. It structures how agents and humans collaborate across a project so that autonomous work stays coordinated and directed rather than ad hoc.
- [Boucle de codage agentique](https://www.thekb.eu/en/glossary/boucle-de-codage-agentique/): The short, minutes-scale loop in which a coding agent works against a product spec and evaluation set: it acts, its output is checked, and it iterates. Treating this loop — rather than a single prompt — as the unit of work is central to reliable agent-driven development.
- [commerce agentique](https://www.thekb.eu/en/glossary/commerce-agentique/): Commerce conducted by AI agents acting on behalf of consumers — searching, comparing, and completing purchases with limited human involvement. Described as an emerging and accelerating category, it shifts the buyer from a person browsing to an agent transacting, reshaping how products are discovered, priced, and sold online.
- [Compaction](https://www.thekb.eu/en/glossary/compaction/): The automatic summarization of a conversation that replaces accumulated history with a condensed version, freeing context space while preserving intent. It is a common countermeasure to context rot, letting long agent sessions continue without the earlier turns crowding out what still matters.
- [Compound Engineering](https://www.thekb.eu/en/glossary/compound-engineering/): An engineering practice in which each unit of work also improves the system that produces future work: lessons, tests, prompts, and tooling are captured as durable artifacts so quality and speed compound over time instead of resetting with each task. Discussed mainly in the context of AI-assisted software development.
- [Compounding Knowledge Lifecycle](https://www.thekb.eu/en/glossary/compounding-knowledge-lifecycle/): A cycle of capture, storage, retrieval, and refresh that makes organizational knowledge composable over time. Each pass adds durable, reusable artifacts so that later work builds on earlier work instead of restarting, echoing the compounding logic applied to teams and engineering.
- [compounding teams](https://www.thekb.eu/en/glossary/compounding-teams/): Teams that no longer write code directly but build recursive frameworks around models — tooling, prompts, and processes that make each future task cheaper and better. The label captures a shift in where engineering effort goes: into the system that produces work, not the work itself.
- [context engineering](https://www.thekb.eu/en/glossary/context-engineering/): The discipline of deliberately assembling, structuring, and pruning the information given to a language model — instructions, retrieved documents, code, and history — so the limited context window carries exactly what a task needs. Treated as a first-class engineering concern, distinct from the wording of a single prompt.
- [Context Flywheel](https://www.thekb.eu/en/glossary/context-flywheel/): The compounding effect by which iteratively curated context improves each successive agent task: better context yields better output, which in turn enriches the context for the next task. It frames context as an asset that accrues value rather than input assembled fresh each time.
- [Context Rot](https://www.thekb.eu/en/glossary/context-rot/): The gradual degradation of a model's output quality as its context window fills with accumulated, partly irrelevant history: earlier instructions get diluted, contradictions creep in, and attention spreads thin. Motivates practices such as compaction, summarization, and starting fresh sessions to keep context dense and relevant.
- [cycle SFEIR à 11 phases](https://www.thekb.eu/en/glossary/cycle-sfeir-a-11-phases/): An AI-driven software development lifecycle organized in eleven phases, numbered 0 to 10, with three human gates and two capitalization points. It formalizes where humans intervene in an otherwise agent-run process, and where knowledge is captured for reuse across future projects.
- [dette technique](https://www.thekb.eu/en/glossary/dette-technique/): The accumulated cost of expedient code and design choices that must later be reworked — interest paid in slower change and more defects. In AI-assisted development it recurs as a specific risk: agents can generate large volumes of code that passes tests yet erodes long-term quality when generation outpaces review.
- [DICE](https://www.thekb.eu/en/glossary/dice/): DICE (Domain-Integrated Context Engineering) — an extension of context engineering that folds an explicit domain model into how inputs and outputs are structured for a language model. By encoding domain rules into the context, it aims to make agent behavior more predictable on specialized tasks.
- [Floating platform](https://www.thekb.eu/en/glossary/floating-platform/): A platform-evolution strategy: as the underlying platform absorbs a capability, the now-redundant custom pieces are discarded and the build point is raised to the next layer of value. It keeps a product riding above commoditized foundations instead of maintaining what the platform already provides.
- [framework 6 étapes](https://www.thekb.eu/en/glossary/framework-6-etapes/): A six-step framework for building AI-assisted solutions: define, design a standard operating procedure, build an MVP, connect, test, and deploy. It gives a repeatable path from intent to a running system, keeping design and verification explicit at each step rather than improvised.
- [GDPval](https://www.thekb.eu/en/glossary/gdpval/): A benchmark measuring AI models on expert-level professional tasks across fields such as finance, law, retail, and software, graded blind by specialists with years of experience. Reported scores reach 40-49% of expert level but require extensive human framing, so the metric itself is seen as under-specified.
- [git worktrees](https://www.thekb.eu/en/glossary/git-worktrees/): A Git feature that checks out several branches into separate working directories from one repository, used to isolate parallel tasks on the same codebase. It lets multiple agents or experiments run side by side without their changes colliding, then merge back independently.
- [Goût développeur](https://www.thekb.eu/en/glossary/gout-developpeur/): A developer's judgment about what a good solution looks like before writing it, together with the discipline to pursue that standard. In AI-assisted work it gains weight: when generation is cheap, knowing which output is worth keeping becomes the scarce, human contribution.
- [grill-with-docs](https://www.thekb.eu/en/glossary/grill-with-docs/): An engineering skill for upstream design, DDD-flavored, that works by a sequential interview guided by four principles — interview, precision of language, evidence, and iteration. It draws out requirements and constraints before code is written, so design decisions rest on stated evidence rather than assumption.
- [Harness](https://www.thekb.eu/en/glossary/harness/): The scaffolding layer around a language model that turns it into a working agent — tools, prompts, memory, execution environment, and safety checks. A common formulation holds that the model is a small part of an agent's practical capability and the harness the larger part, making it a distinct engineering object.
- [Harness engineering](https://www.thekb.eu/en/glossary/harness-engineering/): The building and tuning of the scaffolding around a language model that turns it into a working agent: tools, file access, execution environment, feedback loops, and safety checks. The harness — not the raw model — determines much of an agent's practical capability, making it a distinct engineering activity.
- [infrastructure contexte codifié](https://www.thekb.eu/en/glossary/infrastructure-contexte-codifie/): A three-tier architecture for persistent memory across AI agents, letting context outlive a single session. Reported use spans 283 sessions, 2,801 prompts, and over 16,000 autonomous turns — evidence that durable, codified context can sustain long-running agent work rather than resetting each time.
- [Kishōtenketsu](https://www.thekb.eu/en/glossary/kishotenketsu/): A four-part narrative structure from East Asian tradition — introduction, development, twist, and reconciliation — that builds without relying on conflict. It is invoked as a model for structuring explanation or design, offering an alternative to problem-solution framing when presenting ideas.
- [loi de Goodhart](https://www.thekb.eu/en/glossary/loi-de-goodhart/): Goodhart's law — when a measure becomes a target, it ceases to be a good measure, because optimizing the metric distorts the behavior it was meant to track. It is a caution for AI development: benchmarks and productivity metrics can be gamed once teams steer directly at them.
- [Loop Engineering](https://www.thekb.eu/en/glossary/loop-engineering/): The design and tuning of the iterative cycle an autonomous coding agent runs — plan, act, observe, correct — including how results are fed back, when the loop terminates, and how errors are recovered. The loop, rather than any single prompt, is treated as the core unit to engineer.
- [Orchestration d'agents](https://www.thekb.eu/en/glossary/orchestration-d-agents/): The practice of directing multiple AI agents through a task: chaining them, tracking their state, and recovering when one fails. It requires knowing agents' failure modes to compose them reliably. Practitioners note the open-source layer for registry, lifecycle, permissions, and skills is still largely missing.
- [orchestration multi-agents](https://www.thekb.eu/en/glossary/orchestration-multi-agents/): The coordination of several specialized agents working toward one goal — routing tasks between them, managing shared state, recovering from errors, and running independent work in parallel. Orchestration gains specialization and throughput beyond a single agent, at the cost of added coordination logic. It recurs in complex agent-native workflows.
- [Outcome-based pricing](https://www.thekb.eu/en/glossary/outcome-based-pricing/): A pricing model in which software is billed for the work or results it delivers rather than for seats or licenses. Discussed as a likely shift for AI products, where autonomous agents perform tasks that were previously labor: revenue moves from fixed per-user fees toward the economics of operations and outcomes.
- [Paradoxe de Jevons](https://www.thekb.eu/en/glossary/paradoxe-de-jevons/): An economic observation from William Stanley Jevons (1865): greater efficiency in using a resource can raise total consumption rather than lower it, because falling cost expands demand. Applied to AI-assisted development, it suggests that making code far cheaper to produce may increase — not reduce — the total volume written and maintained.
- [personal software](https://www.thekb.eu/en/glossary/personal-software/): A single-purpose application created by and for one person, fitted to their precise need rather than a general market. Cheap AI generation makes such throwaway, tailored software newly practical, shifting some building from shared products toward disposable tools an individual makes for themselves.
- [Phase Build](https://www.thekb.eu/en/glossary/phase-build/): A two-hour block of autonomous construction in which a candidate builds with AI tools and frameworks of their choice. Used in assessment, it observes how a developer directs agents under time pressure rather than testing recall, foregrounding judgment and workflow over syntax.
- [pipeline de vérification adversariale multi-agents](https://www.thekb.eu/en/glossary/pipeline-de-verification-adversariale-multi-agents/): A verification pattern combining a generator agent, independent reviewers, an automated check (tests or formal methods), and consensus by vote. By pitting agents against one another before accepting a result, it aims to catch errors that a single agent would confidently pass.
- [Plan mode](https://www.thekb.eu/en/glossary/plan-mode/): An agent operating mode that separates planning from execution: the agent first proposes a step-by-step plan for a human to review and approve, and only then carries it out. It reduces wasted or unsafe actions on complex tasks by front-loading intent and human oversight before any change is made.
- [procédure infographique](https://www.thekb.eu/en/glossary/procedure-infographique/): A working procedure built around infographic-quality visual presentation, cited with reference to Steve Jobs's obsession with perfection. It treats the clarity and finish of a visual artifact as part of the method itself, not decoration added after the substance is settled.
- [Programme de tutorat IA](https://www.thekb.eu/en/glossary/programme-de-tutorat-ia/): A structured AI-mentoring program running six weeks, with twelve ninety-minute sessions held twice a week. It formalizes how practitioners are trained to work with AI tools over time, treating adoption as a taught skill rather than something picked up incidentally.
- [PROJ-AI](https://www.thekb.eu/en/glossary/proj-ai/): PROJ-AI — a lightweight methodological layer that makes collective projects transmissible through a repository, an agent, and a shared doctrine. It turns projects into reusable artifacts, so that method and context carry over between teams instead of being rebuilt each time.
- [SDLC](https://www.thekb.eu/en/glossary/sdlc/): The Software Development Life Cycle — the sequence of phases through which software is defined, built, verified, deployed, and maintained. Traditionally designed around human teams and largely invariant, it is the reference frame against which AI-era changes are measured as agents compress, reorder, or automate individual phases.
- [SecNumCloud](https://www.thekb.eu/en/glossary/secnumcloud/): A high-level French security qualification for cloud services, defining the requirements a provider must meet to host sensitive workloads. It functions as a trust and sovereignty benchmark, shaping which platforms are eligible for regulated, public-sector, or otherwise security-critical use in France.
- [skills](https://www.thekb.eu/en/glossary/skills/): A harness primitive that packages reusable agent capability as persistent, shareable files — commonly Markdown (SKILL.md) — loaded on demand. Skills implement progressive disclosure: instructions and tools enter the context only when needed, keeping it dense and guarding against context rot. They make agent behavior testable and portable.
- [Software Factory](https://www.thekb.eu/en/glossary/software-factory/): Non-interactive development driven by specifications and scenarios, without human intervention in the loop — reported at a scale of about $1,000 in tokens per human engineer per day. It frames software production as an automated pipeline where humans set specs and agents execute them.
- [subagents](https://www.thekb.eu/en/glossary/subagents/): Specialized secondary agents that a primary agent spawns to handle a bounded sub-task — searching, reviewing, or transforming — each with its own context and tools. Delegating to subagents keeps the parent agent's context focused and lets independent pieces of work run in parallel.
- [Tension Map](https://www.thekb.eu/en/glossary/tension-map/): A mapping of a market's contradictions and pressure points — rather than its market shares — used to reveal opportunity spaces. By locating where forces pull against each other, it surfaces openings that a share-based view of competition would miss.
- [token](https://www.thekb.eu/en/glossary/token/): The base unit of generative AI processing and cost — a short chunk of text, image, or audio a model reads or produces, roughly a syllable of text. Pricing, context limits, and spend are counted in tokens, which makes it an emerging economic unit: as its cost falls, what is worth generating changes.
- [Usine Logicielle Augmentée](https://www.thekb.eu/en/glossary/usine-logicielle-augmentee/): A software value chain orchestrated by specialized AI agents across six production lines, with human intervention limited to two defined moments. It applies a factory model to software: work flows through agent-run stages, and people act at a few deliberate control points.
- [vibe coding](https://www.thekb.eu/en/glossary/vibe-coding/): A programming style, named by Andrej Karpathy, in which a developer drives an AI agent mostly through natural-language intent and accepts its output without closely reading every line, iterating by feel rather than by manual editing. Effective for prototypes; risky for production code that requires careful review.
- [Vibe Reviewing](https://www.thekb.eu/en/glossary/vibe-reviewing/): Code review assisted by AI agents but validated by rigorous human sign-off. It parallels vibe-coding on the review side: agents surface issues and assessments at speed, while a person keeps final responsibility for what is accepted into the codebase and what is sent back.
- [Wardley Mapping](https://www.thekb.eu/en/glossary/wardley-mapping/): A visual strategy technique that maps a value chain against the evolution of its components, from novel to commoditized. It helps teams see where to build, buy, or outsource, and is applied to reason about positioning in fast-moving AI tooling markets.
- [workflow IA Wardley](https://www.thekb.eu/en/glossary/workflow-ia-wardley/): A workflow that automates the production of Wardley maps with AI assistance, turning strategic mapping from a manual exercise into a repeatable, tool-supported step. It lowers the effort of keeping a value-chain map current as market conditions and component maturity change over time.

## Entities
- [Anthropic](https://www.thekb.eu/en/entities/organization/anthropic/): Organization, 44 fiches
- [Claude Code](https://www.thekb.eu/en/entities/technology/claude-code/): Technology, 53 fiches
- [SFEIR](https://www.thekb.eu/en/entities/organization/sfeir/): Organization, 10 fiches
- [Ethan Mollick](https://www.thekb.eu/en/entities/person/ethan-mollick/): Person, 15 fiches
- [OpenAI](https://www.thekb.eu/en/entities/organization/openai/): Organization, 22 fiches
- [Google](https://www.thekb.eu/en/entities/organization/google/): Organization, 20 fiches
- [Block](https://www.thekb.eu/en/entities/organization/block/): Organization, 10 fiches
- [vibe coding](https://www.thekb.eu/en/entities/methodology/vibe-coding/): Methodology, 22 fiches
- [Mistral AI](https://www.thekb.eu/en/entities/organization/mistral-ai/): Organization, 6 fiches
- [IA](https://www.thekb.eu/en/entities/technology/ia/): Technology, 9 fiches
- [Boris Cherny](https://www.thekb.eu/en/entities/person/boris-cherny/): Person, 11 fiches
- [Buzz](https://www.thekb.eu/en/entities/technology/buzz/): Technology, 7 fiches
- [Arthur Mensch](https://www.thekb.eu/en/entities/person/arthur-mensch/): Person, 3 fiches
- [agents IA](https://www.thekb.eu/en/entities/technology/agents-ia/): Technology, 4 fiches
- [Léon XIV](https://www.thekb.eu/en/entities/person/leon-xiv/): Person, 1 fiches
- [Claude](https://www.thekb.eu/en/entities/technology/claude/): Technology, 11 fiches
- [Agent Client Protocol](https://www.thekb.eu/en/entities/technology/agent-client-protocol/): Technology, 4 fiches
- [Addy Osmani](https://www.thekb.eu/en/entities/person/addy-osmani/): Person, 8 fiches
- [Compound Engineering](https://www.thekb.eu/en/entities/methodology/compound-engineering/): Methodology, 7 fiches
- [MCP](https://www.thekb.eu/en/entities/technology/mcp/): Technology, 14 fiches
- [Cursor](https://www.thekb.eu/en/entities/technology/cursor/): Technology, 14 fiches
- [Claude Skills](https://www.thekb.eu/en/entities/technology/claude-skills/): Technology, 4 fiches
- [Didier Girard](https://www.thekb.eu/en/entities/person/didier-girard/): Person, 4 fiches
- [IA générative](https://www.thekb.eu/en/entities/technology/ia-generative/): Technology, 4 fiches
- [Olivier Rafal](https://www.thekb.eu/en/entities/person/olivier-rafal/): Person, 4 fiches
- [Chris Williams](https://www.thekb.eu/en/entities/person/chris-williams/): Person, 1 fiches
- [Cursor](https://www.thekb.eu/en/entities/organization/cursor/): Organization, 1 fiches
- [Universal Commerce Protocol](https://www.thekb.eu/en/entities/technology/universal-commerce-protocol/): Technology, 4 fiches
- [MCP-UI](https://www.thekb.eu/en/entities/technology/mcp-ui/): Technology, 3 fiches
- [Microsoft](https://www.thekb.eu/en/entities/organization/microsoft/): Organization, 9 fiches

## Agent resources
- [MCP server](https://mcp.thekb.eu/): read-only Model Context Protocol server (11 tools) — docs at https://www.thekb.eu/en/developers/mcp/
- [About & machine surface](https://www.thekb.eu/en/about/)
- [Knowledge graph](https://www.thekb.eu/en/entities/)
- [RSS](https://www.thekb.eu/en/rss.xml)

## Optional
- [Full corpus export](https://www.thekb.eu/llms-full.txt): all 401 fiches in one Markdown file (~1.2 MB)
- [Sitemap](https://www.thekb.eu/sitemap-index.xml)