The new operator stack — five tools that crossed mass adoption in 2025.

What mass adoption actually means in operator tooling

The phrase mass adoption obscures more than it clarifies in the operator-tooling category, because the buyer of each of these tools is rarely the IT department and the procurement process is rarely the IT procurement process. The buyer is the operator. The procurement process is a $20 monthly seat on a corporate card. The adoption pattern is what the SaaS literature calls product-led growth, but applied to a buyer population — chiefs of staff, heads of operations, senior individual contributors, founders — who have effectively unlimited authority to adopt a $20 monthly tool and effectively zero authority to migrate a $200,000 annual contract. The economics of the buyer determine the economics of the seller. Each of the five tools in this analysis built its commercial profile by accepting that constraint and optimising against it.

The crossing point for each tool is defined here against a public metric the company has either disclosed or that has been confirmed by INTELAR through buyer-side procurement records. Granola's crossing point was its one-million-DAU disclosure in November 2025, against the 14,000 DAU it reported at the close of 2023. The compound monthly growth rate across those 23 months works out to 21.3 per cent, which is the highest sustained growth rate for any operator-tooling category INTELAR has measured. Linear's crossing point is harder to define because the company has been growing steadily since 2020, but the inflection in 2025 is clearly identifiable in the migration data: 41 per cent of the year's new paid organisations were migrations from Jira, against 23 per cent in 2024 and 11 per cent in 2023. Linear became the default destination for teams leaving Jira during 2025. Notion AI's crossing point is more diffuse and worth a separate section. Claude Code crossed when Anthropic's internal data — figures that have leaked through procurement conversations and that INTELAR has reconstructed from multiple buyer-side sources — showed that 38 per cent of new Claude API spend in Q3 and Q4 2025 was attributable to Claude Code agents, not to direct API calls. Cursor crossed when the company's $250 million annualised run rate, disclosed in a December 2025 round, made it the highest-revenue AI-native developer tool by an order of magnitude.

The pattern across the five tools is the buyer's response when asked to remove the tool. The chief of staff who used Granola for six months does not return to taking notes manually. The engineering lead who shipped one feature using Claude Code does not return to writing the boilerplate. The team that migrated to Linear does not run a parallel Jira instance during the transition. The data scientist who switched to Notion AI for cross-doc reasoning does not maintain a separate Roam graph for the same work. The IDE switcher to Cursor does not keep VS Code open in a second window. The replacement cost — the writing of the memo explaining why the tool is being removed — is high enough that the buyer's default behaviour is to keep paying. The retention figures show this clearly. Granola's gross monthly churn ran at 1.4 per cent through 2025. Linear's net revenue retention ran at 138 per cent. Claude Code's seat-level churn is not separately reported but the buyer-procurement records suggest a figure below the company's overall retention. Cursor's net revenue retention ran at 162 per cent on the company's own disclosure. The crossing of mass adoption produced a population of users for whom unsubscribing is a strictly worse outcome than continuing to pay. That is what the threshold actually marks.

Granola and the meeting-memory default

Granola's crossing into default status is the most legible story of the five because the category — meeting memory — did not have a default before Granola arrived. Otter, the venture-backed meeting transcription company, had been the volume leader for years but had not crossed into operator-tier adoption. Fireflies, the bootstrapped competitor, had built a stronger SMB business but had not penetrated the enterprise. Both incumbents focused on the transcription substrate. Granola, by Pedregal's published account, focused on the post-meeting artefact — the structured note that an operator would actually use after the meeting was over — and built the transcription only as the input layer to that artefact. The product decision was the commercial decision: a chief of staff who keeps Granola open during a meeting gets back a structured summary that she actually reads, against templates she configures herself. The transcript is incidental. The artefact is the product.

The conversion-from-trial figure is where Granola's adoption story breaks from the SaaS norm. The company's published funnel — disclosed at the SaaStr conference in November 2025 — shows a 31 per cent trial-to-paid conversion at the individual-seat tier. The category baseline is closer to 8 per cent. Granola's conversion is explained partly by the product's instantaneous value — a meeting captured by the tool produces an artefact within five minutes of the meeting ending — and partly by the network effect within an organisation. The chief of staff who adopts Granola begins sharing the artefacts in Slack or via Notion within two weeks of adoption. The colleagues who receive the artefacts adopt the tool within four weeks at a rate of roughly one new seat per two existing seats. The organic seat growth inside a paying organisation runs at 14 per cent monthly on Granola's average through Q3 2025, before the company's enterprise sales motion has even engaged.

Granola's pricing is the question buyers ask about most often. The individual tier is $25 per month, paid annually, with a generous free tier that includes 25 meetings per month. The team tier — which adds shared meeting libraries, template inheritance, and admin controls — is $35 per seat per month. The enterprise tier, which Granola began offering in Q2 2025, is priced at the seller's discretion based on data-residency requirements, dial-in bot support volume, and integration scope. INTELAR's buyer-side records show enterprise contract values ranging from $80,000 to $420,000 annually for organisations between 500 and 5,000 seats. The pricing is unremarkable. The unit economics — gross margin reported in the 78 to 82 per cent range across tiers, customer-acquisition cost recovered in under five months on the team tier — are the more telling figures. Granola is not subsidising adoption. The product is paying its own way at the tier where it sells.

Linear and the execution-graph migration

Linear's adoption story is the longest of the five and the most instructive about what it takes to displace an incumbent in a category that had been considered settled. Jira, the Atlassian-owned issue tracker that had been the default for engineering organisations since the mid-2000s, was the incumbent Linear set out to replace. The replacement did not happen quickly. Linear took its first paying organisation in 2019. By the close of 2022 it had reached 4,200 paying organisations. By the close of 2024 it had reached 11,800. By the close of 2025 it had reached 18,400 — a year in which 6,600 net new organisations adopted the tool, and in which the Jira-migration share of the new cohort hit 41 per cent. The acceleration is real. The mechanism producing it is more interesting than the figure itself.

The mechanism is what Linear's go-to-market team calls the execution-graph extraction. Engineering organisations that migrate from Jira to Linear are not switching tools. They are extracting an opinionated data model — Linear's cycles, projects, and issue graph — from a tool that was structurally agnostic about how the data model was used. Jira allows any team to configure any workflow, which is the source of both its strength and its decline. Linear ships an execution model with the product: cycles are two-week sprints by default, projects are scoped to a defined deliverable, issues belong to a cycle and a project, and the connections are typed. The agent integrations that arrived in 2025 — Devin, Claude Code, Cursor's background agents — read and write against Linear's typed graph in ways that would be ambiguous against Jira's freeform configuration. The agents are easier to integrate with Linear because Linear has decided what the data means. Jira has not.

Linear's pricing reflects the company's commercial discipline. The free tier covers teams of up to 10 members. The Standard tier is $10 per user per month. The Plus tier is $14. The Enterprise tier — which adds SAML, audit logs, and the advanced security controls that buyers above 200 seats require — is priced by negotiation and runs between $24 and $32 per user per month in the buyer records INTELAR has reviewed. Linear's net revenue retention figure of 138 per cent is the headline. The undisclosed figure that buyer-side data exposes is the time-to-expansion: Linear's average paying organisation goes from initial purchase to a seat-count expansion within 90 days of signing. The execution model is contagious within the organisation. Teams that adopt Linear in one function find adjacent functions adopting it within a quarter, and the seat count grows without a renewed sales motion.

Notion AI and the workspace-agent inflection

Notion AI's crossing is the most diffuse and the most contested of the five, because the underlying tool — Notion itself — has been on a long mass-adoption arc since 2018, and the AI layer that Notion has been building on top of the workspace primitive has been an iterative addition rather than a discrete product. The crossing point INTELAR identifies for Notion AI is the early 2025 re-architecture, when Notion shipped what the company calls Agents — long-running, scheduled, cross-workspace reasoning entities that can be configured by an operator and run against the entire workspace corpus without human intervention. The Agents primitive transformed Notion AI from a feature that operators used inside the workspace to a function that operators delegated to the workspace. The buyer's mental model shifted. The retention figures shifted with it.

Notion's published Q4 2025 figures show that 64 per cent of the company's paying organisations had at least one Agent configured against their workspace by year-end. The figure was 19 per cent at the close of Q2 2025, before the Agents primitive shipped to general availability. The growth across two quarters reflects an operator-side adoption that is harder to explain by sales motion alone — Notion's enterprise sales team grew about 30 per cent during the year, but the Agents adoption grew about 240 per cent across the comparable period. The discrepancy is product-led growth at scale. Operators configure their first Agent, see the value, and configure two more within a month. The seat-level engagement data confirms this: the average Agent-using seat at a paying organisation runs 3.4 active Agents by the time the seat is 90 days old. The single-Agent seat is the new-user state. The multi-Agent seat is the default state.

Notion AI's commercial profile sits inside the broader Notion pricing structure, which is what makes the disaggregation hard. The Plus plan is $12 per user per month and includes a basic AI allowance. The Business plan is $18 and includes the Agents primitive. The Enterprise plan is negotiated. The disclosed AI-attributable ARR figure is the one Notion has been least willing to publish, and the buyer-side records INTELAR has reviewed produce a wide range — between 18 and 34 per cent of the Business and Enterprise tier ARR, depending on how the buyer attributes the seat upgrade from Plus to Business. The methodology question is not trivial. Notion's commercial position is that the Agents primitive is what's pulling buyers from Plus to Business, and the AI-attributable share of revenue is therefore the larger figure. The buyer's position is sometimes that the Business tier was always going to be the next step. Both are partially right. The question for the category is whether Notion AI is a feature of Notion or a separately defensible product, and the answer the market is still working out is closer to the former than to the latter — which has commercial implications that the next section of this article does not have space to fully cover.

Claude Code and the engineering-tool default

Claude Code's crossing is the most recent and the steepest. Anthropic shipped Claude Code to public availability in February 2025. The tool is a command-line agent that takes a coding task and executes it across the user's codebase with full read-write access. The product category is contested — Cursor positions itself in adjacent territory, GitHub Copilot positions itself differently, and a longer tail of agent-coding products has been multiplying since mid-2024 — but Claude Code is the tool engineering organisations have been adopting fastest as the agent-coding default. Anthropic's published figure of 7,200 organisations with active Claude Code seats at the close of 2025 represents about 18 per cent of the company's overall organisation count, but the disproportionate share of the API spend — 38 per cent of new Claude API revenue in the second half of 2025 — indicates that the seats that exist are running heavy.

The conversion data for Claude Code differs from the other four tools in this analysis because the tool is not sold on a seat basis. The pricing model is metered API consumption with a tiered organisation-level commit, and the conversion question becomes: what proportion of engineers in an organisation actually run Claude Code daily, and what is the per-engineer spend at steady state. INTELAR's buyer-side records produce a range that does not converge cleanly. An organisation with 200 engineers and a fully rolled-out Claude Code adoption runs a monthly API bill between $180 and $640 per active engineer, depending on the codebase and the team's task profile. The same organisation, before Claude Code, ran Anthropic's API at less than $40 per engineer per month for all uses combined. The per-engineer spend has increased an order of magnitude. The output has scaled with it: engineering organisations running Claude Code at full adoption report between 22 and 41 per cent productivity gains on the company-published figures, with the variance explained by codebase quality, eval discipline, and the team's pre-existing automation maturity.

The operator persona driving Claude Code adoption is not the typical engineering procurement persona. The chief technology officer matters in the procurement decision, but the actual adoption motion runs through the senior engineers who try the tool on their own projects and the engineering managers who see the output and commit the team. The bottoms-up adoption pattern is structurally similar to Granola's, with the difference that the API spend at full adoption is large enough to require a formal procurement conversation. The procurement conversation is happening more often. Anthropic's enterprise sales motion grew about 60 per cent in 2025 to handle the volume, and the buyer-side records suggest that the procurement conversation now closes faster — the median time from initial conversation to signed paper has compressed from 71 days at the close of 2024 to 38 days at the close of 2025. The tool has become a known quantity in engineering procurement. The conversation is no longer about whether agent-coding works. It is about which tool and at what commit. Claude Code is winning the answer.

Cursor and the IDE shift

Cursor's crossing is the most commercially visible of the five because the company has been the most willing to disclose its run-rate figures. The $250 million annualised revenue disclosure in December 2025 made Cursor the highest-revenue AI-native developer tool by a meaningful margin, and the 480,000 paid seats figure produced a per-seat revenue average around $520 annually that fits the company's mid-tier pricing of $20 to $40 monthly across plans. The growth profile is steep: Cursor's annualised revenue was reportedly under $4 million at the close of 2023 and around $65 million at the close of 2024. The 2025 growth rate exceeded any comparable developer-tooling category since GitHub's earliest expansion phase. The pattern is one Anthropic, OpenAI, and the broader model layer all have a commercial stake in, because Cursor's underlying inference spend — which the company has disclosed runs at roughly 35 per cent of revenue — is the largest single revenue line item for the model providers.

The product positioning that produced the adoption is harder to copy than it looks. Cursor is an AI-first IDE built as a fork of VS Code, which means the migration cost from VS Code is effectively zero — the engineer opens Cursor, signs in with their existing GitHub credentials, and the IDE inherits their settings, their extensions, and their layout. The friction that has historically prevented IDE adoption is gone. The agent capabilities — the chat interface, the inline edit suggestions, the background agents that take multi-step tasks against the codebase — are an addition rather than a replacement. An engineer can adopt Cursor and use it like VS Code for a week before exercising any of the AI features. By the end of that week, the inline suggestions have produced enough value that the engineer engages the chat interface; by the end of the second week, the engineer is running background agents. The adoption curve is gentle. The terminal state is dependent on the AI features.

Cursor's enterprise commercial profile is the place where the company's product-led adoption is hitting its first procurement-side test. The buyer-side records INTELAR has reviewed show that Cursor's enterprise contracts in the second half of 2025 ranged from $140,000 to $1.8 million annually, with the upper end representing organisations of 2,000-plus engineers on the Cursor Business tier with custom data-residency and audit-log requirements. The Business tier is priced at $40 per user per month before enterprise customisation. The unit economics work because the inference cost is passed through with a margin that has narrowed under model-provider pricing pressure but has not yet inverted. The strategic question for Cursor — and for the model providers that depend on Cursor's volume — is whether the inference-pass-through model can hold its margin as the next generation of base models drives down the cost of the most expensive call patterns. The next two quarters will produce the answer. INTELAR will be tracking it.

What to watch

The five tools that crossed mass adoption in 2025 will face their first round of structural pressure in 2026 — pricing competition, model-provider strategic posture, and the second-generation entrants that are now visible in the venture pipeline. The category will not look the same at the close of the year.

• Whether Granola's enterprise sales motion compresses the per-seat economics that have driven the company's compound monthly growth; the team-tier price point of $35 per seat assumes a low support burden, and the enterprise contracts INTELAR has reviewed run at a higher support cost that is not yet fully reflected in the company's gross margin.
• Whether Linear's 41 per cent Jira-migration share holds against Atlassian's strategic response; Atlassian's Rovo agent layer has begun closing the integration gap with Linear, and the question is whether Atlassian's incumbency in regulated industries — banking, healthcare, government — will hold the buyers that have not yet migrated.
• Whether Notion AI's Agents primitive scales beyond the workspace-resident use case; the company's H1 2026 roadmap includes external connectors that would extend Agent reach into Slack, GitHub, Linear, and the broader operator stack, and the integration depth will determine whether Notion AI is competing with Microsoft Copilot or with the broader category of cross-tool agents.
• Whether Claude Code's API-spend profile remains the largest single revenue line for Anthropic's enterprise tier; the next generation of Claude models is expected to drive down the per-task cost, and the question is whether Claude Code's volume can scale fast enough to compensate for the cost compression, or whether Anthropic will need to introduce a per-seat pricing layer that the market has not yet seen.
• Whether Cursor's inference-pass-through margin holds against the model-provider pricing pressure; the company's gross margin has narrowed about 600 basis points across 2025 as model providers have responded to demand-side leverage, and the strategic question is whether Cursor can build a sufficiently differentiated product layer to defend a margin that is no longer protected by inference scarcity alone.

Frequently asked

What does it mean for an operator tool to cross from niche to default?

Three observable signals. The operator who adopted the tool stops describing it as a tool and starts describing it as a function — Granola becomes meeting memory, not Granola. The trial-to-paid conversion breaks above the category baseline by a meaningful multiple. And the buyer who replaces the tool has to write an internal memo explaining why. The five tools in this analysis cleared all three tests during 2025. The 2026 question is whether the crossing produces sustained pricing power or whether the category attracts the next generation of competitors that compress the incumbents' margins.

How does Granola's growth profile compare to the meeting-memory category?

Granola moved from approximately 14,000 daily active users at the close of 2023 to more than one million at the November 2025 disclosure. The compound monthly growth rate across the 23-month window is 21.3 per cent, which is the highest sustained growth rate INTELAR has measured in any operator-tooling category. The category incumbents — Otter, Fireflies — grew over the same period at single-digit monthly rates. The shift in the category is what economists call a regime change rather than a competitive shift: the buyers are not switching tools; the buyers are entering a category they had not previously been buying in.

Why did Linear's Jira-migration share accelerate so sharply in 2025?

Two reinforcing reasons. The execution-graph data model Linear ships — typed connections between cycles, projects, and issues — is materially easier for agent integrations to read and write against than Jira's freeform configuration. The agent integrations that arrived in 2025, including Devin, Claude Code, and Cursor's background agents, all integrated more cleanly with Linear. The second reason is the cost of staying on Jira: as engineering teams begin running more agent-driven work, the configuration overhead of Jira's flexibility becomes a tax on the agent workflows. Linear is the lower-tax destination, and the migration economics started reflecting that reality in 2025.

Is Notion AI a separately defensible product or a feature of Notion?

The question the category is still working out. The Agents primitive that shipped in early 2025 has the strongest case for product-level defensibility: it is configured by the operator, runs scheduled work against the workspace corpus, and integrates with external systems through connectors. The case against treating Notion AI as separately defensible is that the value of the Agents primitive is downstream of the workspace's structural completeness — an Agent against a sparse workspace produces sparse output. The right answer for the buyer is to evaluate the bundled price against the bundled value, which is exactly what Notion's commercial model assumes.

What is the actual procurement profile for Claude Code at the enterprise tier?

The procurement profile is API-consumption with an organisation-level commit. The buyer typically signs a 12-month commit at a discounted per-token rate, in exchange for predictable API capacity and an enterprise support tier. The active-engineer monthly spend ranges between $180 and $640 in INTELAR's buyer records, with the variance explained by codebase scale and task profile. The procurement conversation has compressed substantially across 2025 — median time from initial conversation to signed paper has moved from 71 days to 38 days — which indicates that engineering procurement teams have become comfortable enough with agent-coding as a category that the diligence cycle has shortened.

Will Cursor's inference-pass-through margin survive the next model-pricing cycle?

The honest answer is unclear. The company's gross margin has narrowed about 600 basis points across 2025 as model providers responded to demand-side leverage. The narrowing has not been catastrophic because Cursor's revenue scaled faster than the margin compressed. The next cycle's question is whether Cursor can build product differentiation — the background-agent execution layer, the codebase-context retrieval system, the multi-model routing — that justifies a margin not protected by inference scarcity. The competition's answer will arrive when the second-generation AI-first IDEs reach a comparable feature surface, which the venture pipeline suggests is closer than the market currently models.