Build1 distinct publisher3 min readPublished
Microsoft's 18 August cleanup retired the free Deep Research and left its successor inside a paid plan. Any workflow that was generating sourced reports for nothing now needs a budget line or a rebuild.
The Engineer · Build desk
product
Model choice is becoming a line item, and the differentiator moved up the stack1 distinct publisher
build
GitVenom dressed hundreds of repositories over several years to ship AsyncRAT and Quasar1 distinct publisher
build
Microsoft ships an MIT-licensed agent kernel: policy rings, Ed25519 identity, kill switch1 distinct publisher
product
Microsoft never announced a China exit. Five years of filings did it instead1 distinct publisher
Compiled by The EngineerSomething wrong?How this is made
Microsoft's support page draws its line at creation, not access. The quoted sentence is that Microsoft 365 Premium subscribers can continue creating detailed reports and analyses using Researcher in Copilot [5]. The read rights granted elsewhere on that page are broader, so the entitlement boundary runs between opening an old report and producing a new one [16]. On the free tier, the ability to run new deep research is gone [6]. If what you built was a generator, keeping the ability to open an old archive does not bring that generator back.
The arithmetic is worth writing down before someone asks for it in a planning meeting. At the reported monthly figure, a seat costs $239.88 a year [14]. Five people who all leaned on the free tool come to $1,199.40 a year [15]. For that to be the true cost, seats have to map to people rather than to concurrent jobs, and Researcher has to do what Deep Research did.
Retention is where the design is most legible. Group Chat threads, messages and the images created inside those chats were not carried forward past 18 August [11]. Reports and podcasts each got a different exit path. The three features sat inside one app, but they ended up with three different retention outcomes, and what mattered in the end was whether the output could leave in bulk.
Some of this genuinely is housekeeping, and it is fair to say so. TechCrunch's read, quoted in the dev.to piece, is that Microsoft is ditching a number of unsuccessful AI features [8]. Mico, the floating animated character reviewers had already compared to an AI-era Clippy, is the one item on the list whose removal needs no risk assessment [12]. Free chat and image generation survive [3], which covers the casual user and does nothing for someone whose value came from the multi-step sourced report. The dev.to author's own summary is that the change is partly sensible housekeeping and partly a paywall [13]. In my context the second half is the only half with a cost code attached.
Which leaves the condition I would test before signing anything. The documentation names Researcher as the way Premium subscribers keep creating reports, but it makes no comparison between the two tools' output [17]. That gap is the finding: Microsoft is describing a replacement, not measuring one. I would take the three hardest prompts I used to send to Deep Research, run them through Researcher, and read both reports side by side before buying seats for a team. It is a cheap test. It is also the test that shows whether you are buying back the workflow or just buying back the menu item.
Ranked by verification strength, evidence, and original report placement.
As of 18 August, Microsoft began retiring a set of free consumer Copilot features, the standout being Deep Research, the tool that turned a prompt into a long, sourced report, while folding its two Copilot apps into one.
Per Microsoft's own support documentation, the consumer features going away by 18 August are Group Chat, Podcasts, Deep Research, Copilot Labs and Mico.
Microsoft's support page states: "Microsoft 365 Premium subscribers can continue creating detailed reports and analyses using Researcher in Copilot."
For everyone other than Microsoft 365 Premium subscribers, the ability to run new deep research is simply gone.
Microsoft is combining the consumer-facing Copilot app and the Microsoft 365 Copilot business app into a single application, with mobile and web already unified and the Windows and Mac desktop apps following in mid-September.
Distinct publishers with included, body-backed reporting in this cluster.
dev.to
1 article · August 29, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Microsoft's own words, one outlet's price tag
Almost everything consequential here is a direct quotation from Microsoft's support pages — the Researcher entitlement sentence, 'will not be carried forward', the instruction to download podcasts one at a time. For a deprecation story that is close to primary sourcing, and dev.to is careful to mark which lines are Microsoft's. The soft plank is the number in the headline: $19.99 appears as a plan 'reported at' that figure, with no pricing page and nobody credited, and the annual seat maths inherits that softness rather than fixing it.
Live everywhere, counted nowhere
This is already in force, on a date, for every consumer Copilot user who did not choose it: mobile and web are merged, desktop follows in mid-September, and the retired features stopped working on 18 August. That is deployment, not demand. Nothing in this reporting says how many people ran Deep Research, how many podcast libraries went unexported, or whether a single free user has since bought a Premium seat — so the half of adoption that would tell you whether the paywall works is invisible.
Careful prose, one uncertain number
dev.to hedges where a harsher piece would not: it credits the merge as a real simplification, says outright that this is no scandal, and notes that existing reports survive on cheaper plans. The overreach is narrow but load-carrying in the wrong place — an unconfirmed monthly price becomes annual seat maths and a headline, and one consumer-tier retirement is read as the clean case of an industry-wide slide from free to paid.
A conversion funnel, described by the AI downside desk
Microsoft's interest is visible in the shape of the cut: the free feature that did the most valuable work is the one that reappeared inside a paid plan, while chat and image generation stay free as the way in. The other side of the page has a lean too — this account runs under a banner devoted to AI's downside, so the paywall reading is house style. It shows in the framing rather than the facts, since the facts are mostly Microsoft's own sentences.
Trust the documentation, hold the dollar sign
Two different confidences are bundled together here. What happened, and what you lost if you missed 18 August, is quoted from the vendor and safe to act on. What it costs to get the capability back, and whether the replacement is as good, are not established: one unattributed price, no capability comparison, and no second outlet to check either against.