A new approach to answer-engine optimization (AEO) uses a site's own chatbot as a continuous integration test, with the rule that the bot gets nothing the outside world does not get. The agent for Elj…
GitHub has launched stacked pull requests in public preview, a workflow that organizes large changes into an ordered series of small PRs with automatic rebasing and retargeting. The feature is availab…
A Cloudflare study measuring agent-readiness capabilities found that well-known files for agent discovery (llms.txt, sitemap.xml, MCP server cards) were consulted zero times by every AI model tested, …
A new AEO Bench study from an unnamed research group found that in 900 agent tasks across five AI models, none consulted the llms.txt file, sitemap.xml, or any other discovery file when navigating a 4…
A controlled study of Cloudflare's agent-readiness techniques found that no AI model fetched llms.txt or sitemap.xml in 900 test cells, and all models except gpt-oss-120b scored perfect 28 of 28 on pr…
A new benchmark called AEO Bench, built by an unnamed team, measures how AI agents actually interact with site-side readiness techniques such as llms.txt files, sitemaps, and markdown content negotiat…
A four-study benchmark series by Cloudflare's AEO Bench found that no AI model consulted well-known files such as llms.txt, sitemap.xml, or MCP server cards unprompted across 2,900 agent runs, even at…
A study backfill found that open-weight model gpt-oss-120b passed all 13 guardrail gates for $0.33 in API spend, while frontier models moonshotai/kimi-k3 and anthropic/claude-fable-5 both failed the s…
The open-weight model gpt-oss-120b passed all thirteen safety gates in a benchmark series that previously tested only closed-weight models, costing just $0.32 to evaluate. The benchmark, designed to t…
A powered measurement of the client-prune fence across four models found it works perfectly on sonnet (prune cells fell from 16/40 to 0/40, p = .00002) but is unnecessary on fable-5, which spontaneous…
A study by AP found that a guidance sentence instructing models to return an empty tag list for wholly-foreign topics achieved 12/12 accuracy across all three tested models (Opus, Gemini, Sonnet) with…
Cloudflare launched isitagentready.com, a tool that grades websites on how well they serve AI agents across discoverability, content accessibility, bot access control, and capabilities. Only 78 percen…
Cloudflare reported this spring that only 78 percent of websites have a robots.txt and fewer than fifteen sites on the entire internet have implemented some newer standards, prompting the company to l…
Typography-led websites will become the recognized look of premium web design as AI-generated imagery collapses the costly signal that once made image-heavy sites feel high-end, according to a design …
A study by AO found that without guidance, AI models never matched exact canonical casing on a 40-tag fixture catalog across 36 drafting tasks, with 0/36 clean on all three tiers. With a canonical-tag…
Lightning Jar declares that cold email is spam, arguing that the recipient's experience—not the sender's intent—defines spam, and that personalization does not create a relationship. The company refus…
OpenCode co-founder Dax Raad told Syntax podcast hosts Wes Bos and Scott Tolinski that his team has become dependent on a proprietary AI model they are contractually unable to discuss publicly, highli…
A study by AN found that providing a get_template_view tool eliminated the visibility-clause tax in outline views, raising anaphora success to 45–48 out of 48 with near-zero asks across four arms, thr…
A study by an unnamed team building LLM agents found that giving a model a tool to fetch missing information eliminated a 70% refusal tax on document-editing tasks, while adding a prompt rule to prefe…
Sarah Guo, who became Greylock's youngest general partner at 28 and left four years later to found Conviction, staked her firm entirely on AI before ChatGPT launched, with pre-ChatGPT seed checks into…