Sadly librarians who are thirsty for respect after years ignored by faculty, and dismissed by vendors are sometimes willing to overlook serious product deficiencies in exchange for being treated as valued intellectual partners by vendors. The bar for earning that loyalty can be surprisingly low

[Read] Can BM25 be a probability? softwaredoug.com/blog/2026/03...

Undermind.ai has now added a new Projects feature. This now only allows you to create workspaces for your projects, but also make more agentic workflows possible (1)

Looking at a utterly delusional librarian substack post and thinking whether I want to respond. There's a lot I agree with the piece e.g. in terms of the trust users have with the library and how often we see things other institution units may not notice. But equally there's this delusional view (1)

Worth spending time with this article from Scott Cunningham: Summary and my thoughts here: substack.com/@pubtechrada... Original article here: substack.com/home/post/p-...

The more i play with Claude code to combine sources to create the ultimate search agent the more i realise the wisdom of traditional systematic review. Why? (1)

Very much looking forward to contributing to this webinar, and delighted to share the stage with two great colleagues, @katiecorker.bsky.social and Kathleen Shearer! My contribution will be about @barcelonadori.bsky.social. @cwtsnl.bsky.social

Claude Code - subagents vs agent teams - playing with this for literature review tasks. code.claude.com/docs/en/agen... (1)

sigh the thing about writing on cutting edge things is my blog posts get outdated within 6 months if not faster

Looking forward to FORCE2026 keynote- Surviving the Disruption: Scholarly Communication in the Age of AI by Ian Mulvany, Chief Technology Officer, BMJ Group event.fourwaves.com/force2026/sc... - Early Bird Registration for FORCE 2026, 3-5 Jun,Singapore is still available! @ianmulvany.bsky.social

I really think things are speeding up. Let's not even talk about agentic ai. The fact that you can easily ask ai to vibe code basic things on the fly is a game changer for someone like me. I was reading a paper suggesting a stopping rule for screening & I was curious to see how well it worked (1)

I was talking to a young faculty who is a very early adopter of Claude code and is automating a lot of research tasks and he was asking for advice about combining mcp servers/skills for literature review. He quickly realised the problem isnt technical or coding these days but workflow (1)

elicit.com/blog/elicit-... @elicit.com releases their api, yet another tool you can easily integrate into your custom research agent

spent the last 2 weeks really looking at the openalex topic hierachy - domain, field, subfield, topic vs ASJC subject area, main category and subcategory. Can see how openalex first 3 levels, domain , field, subfield is sort of modelled on elsevier's ASJC at least in terms of names. (1)

[Blogged] The agentic researcher - building custom, transparent and extensible workflows with Claude & MCP aarontay.substack.com/p/creating-y...

I really need to study this agentic stuff more

Does clarivate understand what citations are for? "With Nexus, Clarivate are essentially integrity-washing synthetic text, giving it an academic sheen without any academic rigour." "... it offers to "Find Verified Alternative ... (1) www.hughrundle.net/does-clariva...

How do researchers actually use AI-powered science tools? Lessons from 250,000+ queries February 27, 2026 allenai.org/blog/asta-in... from AI2 tools AI2 paperfinder and ScholarQA now under asta. This part caught my eye (1)

Ah looks like scite is officially in the chatgpt app store. No need for the scary developer mode.

Scite officially launches MCP server with support for info about institution full text entitlements via getFTR, libkey. - suspect this might lead to a rush of academic ai search mcp servers? scite.ai/blog/introdu...

Scite officially launches MCP server with support for info about institution full text entitlements via getFTR, libkey. - suspect this might lead to a rush of academic ai search mcp servers? scite.ai/blog/introdu...

I am biased as I helped with evaluation of FORCE 2026 proposals - but the programme looks really good with 3 parallel tracks with really top notch speakers on everything from responsible research assessment to impact of ai on research and publishing workflows(1) share.google/CidMfxRgXCH4...

Watching third iron - libkey strategy with AI tools... Very important topic... wonder how this works with GetFTR if at all..

[Blogged] The agentic researcher - building custom, transparent and extensible workflows with Claude & MCP aarontay.substack.com/p/creating-y...

Reading a pretty nice librarian comparison test of Primo RA, Scopus AI, web of science RA etc. It finds WOS RA is least favoured which tracks (retrieval was quite bad at that time). That said its definition of RAG as something that uses vector embedding for retrieval is wrong..

blog.openalex.org/openalex-api... Might break my vibe coded dashboards and mcp servers because you now need a api key

I've read this paper before. Its a very rigorous method. Sure there's a tiny issue with the fact it doesn't handle the multiple testing issue but ranked quasi sampling seems conservative enough to counter that. I'm curious ai deep search tools dont use this to estimate when to stop. Undermind (1)

[Read] Don't Stop Me Now, `Cause I'm Having a Good Time Screening: Evaluation of Stopping Methods for Safe Use of Priority Screening in Systematic Reviews onlinelibrary.wiley.com/doi/10.1002/... tests 15 statistical/rule-based (heuristic) stopping method, only 1 reliably meets set recall target(1)

Talking to faculty about AI assistants in academic databases  share.google/Vh4y3SE8il6s.... I really like this diagram - very good way to think abt what "AI powered" search means

MEASUREMENT - This article describes measurement of OA materials - a challenge as usage and provision changes katinamagazine.org/content/arti... 2020 saw temp removal of paywalls during the COVID-19 lockdown & the introduction of R&P agreements. Article considers these and identifies usage trends

I notice this pattern. Whenever a new type of tool or technology emerges, naysayers will say its can't ever be useful cos it lacks this property X that the older tech has, despite the fact that millions are using it. And as the technology matures usage increases further & die hards still dig in.(1)

Being hearing from researchers and librarians about the amazing things they can do with Claude code. So much for the "stochastic parrots" cannot be useful no matter what arguments .

For most definitions of grey literature preprints count as GL. But preprints are increasingly a "first class entity" (with dois) easily findable thanks to sources like Semantic Scholar/openalex used by most Ai search tools like Elicit/consensus/undermind etc. (1)

To add on they may be vulnerable to data voids same as Google ..eg there's only a few pages on Topic x and its pretraining also has nothing..

"The AI can access articles from Elsevier’s own collection and those of its four partners: Emerald, the Institute of Physics, the New England Journal of Medicine Group, and Sage Publications". Interesting Wiley isnt on the list. Based on partnership agreements with tools like Scite, Consensus (1)

Only Consensus MCP server is limited to only top 3 results consensus.app/home/mcp/

managed to setup chatgpt with remote MCP servers to search scite, Consensus.app and Pubmed. Remote means you don't have to install anything on your desktop. Needs to turn on developer model though.

[Watched] Finally watched this CNI talk from end to end. This pretty much sums up the cutting-edge issues regarding library + ai discovery. Library nexus, library workflow, library MCP servers. Also excellent q&a discussion youtu.be/SgFsxlCQvOE?...

[Watched] youtu.be/oXqJQeHWhUs?... - good refresher on doing systematic review with undermind.ai, 2dsearch, polygon - I must say many parts even free version of Undermind was heavily praised . Eg the email alert, lack of hallucinations, positive reviews by librarians etc. Not surprising I guess

Even librarian have half a dozen definitions of what library databases or search engine mean. Eg i saw a systematic review librarians suggest database is say medline but platform/interface eg ovid, pubmed, Ebscohost, ProQuest, web of science is "search engine"

The general sense I get is Claude currently is best at searching then Chatgpt with Gemini the worst at least for the non deep research versions.

"On our API, Claude’s web search and fetch tools now automatically write and execute code to filter and process search results, keeping only relevant content in context—improving both response quality and token efficiency." This is going to be very powerful for academic search engines using claude

Curious about papers that investigated how good openalex was when used solely as database for systematic review. Undermind as usual found the top 2 super relevant, 1 that is closely relevant. Claude research mode found 2, OpenAlex deep research 1, Gemini Deep research 1, Consensus Deep search 0

[Watched] Using AI for systematic review: 2025 vs 2026 - interesting point PRISMA 2020 is challenged or constrained by new AI tools. Step by step approach not appropriate for newer methods eg AI can extract PICO from title/abstact, is that screening or "data extraction" ? youtu.be/3ctb-Icri0o?...

Explains a lot

"How do we conceptualise equity, liberty, what counts as progress, what kind of economic growth, and whose knowledge counts as legitimate knowledge? These are not abstract. They are woven into the system. Into funding, journals, rankings, intellectual life, and scientific methods themselves."

Studied a bit more on how Claude skill works support.claude.com/en/articles/... - think we will be seeing a lot of librarians learning this soon

[Blogged] Are AI Tools Killing Review Articles? Two Failure Modes Suggest Otherwise aarontay.substack.com/p/are-ai-too...

For example, one exercise I use is connecting to their own Zotero library through MCP and building different interfaces to explore it and find what works for their goals. Then it's easy to build a reusable skill for their preferred workflow. (2/2)

The only thing close to this is SciSpace agents - which has many academic specific tools connected & it reasons what to do/use but its extremely expensive credit wise & probably not worth it. The DIY approach with setting up MCP servers + Claude Skills is still quite tricky currently(1)