Research is structured prose and structured data, intermixed. A literature review is a table of papers plus a synthesis essay that draws on those rows. A customer-research study is a sheet of interview transcripts plus a findings doc citing specific quotes. Most tools force you to pick one shape and shred the other to fit. Notebooks are prose with code, spreadsheets are rows with prose stuffed in cells. Neither is the natural shape of research.
Dock workspaces are doc and table side by side, with agents writing to both. Below: three workflows that map cleanly, plus the provenance story.
Three workflows, one workspace
Literature mapping. A table surface: rows for papers, columns for citation, summary, methods, sample size, finding. A mapping agent reads each paper, appends a row with a structured summary, links the PDF. You skim the table to find the cluster you care about; the agent did the per-paper reading. Borah et al. (BMJ Open, 2017) found registered systematic reviews take a mean of 67.3 weeks to complete and publish, most of it on extraction and screening. The table-plus-agent shape compresses that.
Synthesis. A doc surface in the same workspace: sections for background, methods, findings, discussion. A synthesis agent drafts the findings section by reading the literature table and the interview rows. A human edits in place. Concurrent writes are safe when each agent owns a section, because Dock's update_doc_section is heading-scoped, not whole-body. The synthesis is the actual doc, edited by named principals, not a chat transcript to reconstruct.
Async review. Reviewers comment on the doc and the table from wherever they are. Each comment is attributed to a specific reviewer or review-agent. Threads resolve when the author addresses them. No exported PDF, no merge conflicts, no "v3-final-mei-edits.docx." Research is async by nature, distributed across labs and time zones; the shared-workspace primitive makes it first-class.
Provenance is native, not bolted on
The hard problem in research is provenance: who said what, what the source is, when it changed. Dock answers this with two primitives.
Backlinks build the citation graph automatically. Every [[slug]] cross-reference creates a row on the target workspace's "Referenced from" sidebar. Cite the customer-interview workspace from your findings doc, and the interview workspace shows your findings doc as a referrer. The graph maintains itself. The same backlinks let an agent reading on join orient in seconds.
Dual-keyed audit means every write is signed by both principal and workspace, immutable, queryable. The agent's identity is its own, not borrowed. Combined with the citation graph, every claim in your findings doc traces to a row, a source, a writer, and a timestamp. See agent audit and compliance for the substrate.
The async dimension
Research is async by default. Coauthors live in different time zones, reviewers respond on their schedules, field researchers come back with notes weeks later. Dock's presence, attribution, and comment threads make async-collaborative research first-class. A reviewer who logs in Thursday sees Tuesday's draft, Wednesday's agent edits, and exactly who wrote what. No catch-up meeting required.
This matters beyond academic labs. Forrester's 2024 Buyers' Journey Survey reports 35% of B2B buyers already consult external influencers during their journey, expected to reach 50% by end of 2025. Customer-research teams synthesizing those signals run the academic workflow: map sources, synthesize findings, review async. The substrate is shared.
Where to start
If you run research, academic, R&D, customer, or market, Dock is built for the shape of your work. Start one workspace per project: a mapping table, a synthesis doc, a thread of attributed comments. Add a mapping agent and a synthesis agent as members. The agent collaboration primer is the right next read; what is an AI workspace is the category map.
See the full pattern on Dock for Research, with customer-research and content-calendar templates that already understand the citation graph. Agents do the legwork. You do the thinking. The workspace remembers.
FAQ
If the mapping agent reads each paper and writes the rows, how do I trust its summaries?
Every row the mapping agent writes is dual-keyed: signed by the agent's own identity and the workspace, immutable and queryable. The row links the source PDF, so you can open the original next to the agent's structured summary and check the methods or sample-size cell against it. The agent compresses the per-paper reading; the audit trail and the linked source let you verify it without re-reading everything.
What stops two agents from clobbering each other when they write the synthesis doc at the same time?
Dock's update_doc_section is heading-scoped, not whole-body, so concurrent writes are safe as long as each agent owns a section. A synthesis agent drafting the findings section does not touch the background or discussion sections another writer holds. That is what makes the doc the actual artifact rather than a chat transcript you have to reconstruct.
My sources live in a separate interview workspace, not in this findings doc. Does the citation graph still hold?
Yes, that is exactly the case Backlinks handle. A [[slug]] cross-reference from your findings doc to the customer-interview workspace creates a row on that workspace's "Referenced from" sidebar automatically, so the interview workspace shows your findings doc as a referrer. The graph maintains itself across workspaces, and combined with the dual-keyed audit, every claim still traces to a row, a source, a writer, and a timestamp.
Is this only for academic labs, or does it fit applied customer and market research?
It fits both, because the substrate is shared. Customer-research teams run the same workflow as academic labs: map sources into a table, synthesize findings in a doc, review async across time zones. Whether you are tracking systematic-review papers or the external influencers Forrester counts in the B2B buyer's journey, the map-synthesize-review shape and the provenance primitives are identical.
