Reference

Work out which growth curve you are actually on.

Four growth archetypes, the metrics that follow from each, and what the four Ps look like inside them.

From the summary: “Your job is not to force the company onto the Supernova curve. It is to work out which growth curve you are actually on and build the system that curve needs.” Back to the summary

Two of these names come from Bessemer Venture Partners’ State of AI 2025, which supplies the trajectory and margin figures below. Bessemer’s report does not assign companies to either archetype; the company placements throughout this report are mine. Compounder and Cautionary Tale are my own labels.

The four, side by side

	Supernova	Shooting Star	Compounder	Cautionary Tale
To $100M ARR	~1.5 years	~4 years	5+ years	Durable $100M not established
Year 1	~$40M	~$3M	$1M–$3M	Fast, then flat
Year 2	~$125M	~$12M	Doubling	Stall or retraction
Gross margin	~25%, often negative	~60%	60–80%	Negative or unreported
ARR per employee	>$1M	~$164K	Varies	Often the headline stat
Retention	Unproven at speed	Durable	Durable	Weak or undisclosed
My examples	Lovable, Cursor, ElevenLabs, Sierra	Harvey, Glean, Clay, Abridge	Synthesia, Rilla, Photoroom, Writer	Jasper, Bolt.new, 11x, Cluely

Bessemer’s cohorts show average ARR per employee differing by roughly sevenfold. That operating difference drives much of what follows.

The Cautionary Tale examples may be doing well today. For several, I could not find recent public updates with hard revenue, margin, or retention numbers. The silence may reflect a reaction to earlier overstatement rather than poor performance; the public record does not let us know. The label describes evidence gaps and warning signs, not a claim that the companies have failed.

Supernova

Bessemer’s Supernovas are defined by unprecedented growth and adoption: its cohort averaged roughly $40M ARR in year one and $125M in year two, with about 25% gross margin and more than $1M ARR per employee. In this dataset, companies reached that growth curve through both self-serve adoption and unusually fast enterprise sales.

Where it breaks. Margin and retention. Bessemer’s own figure is 25% gross margin, often negative. Speed without improving economics or durable customer behavior can turn a Supernova into a cautionary case. Cursor and ElevenLabs are two examples with continued public acceleration beyond $100M; Bolt.new stopped publishing revenue updates after $40M.

Metrics that follow. Activation rate, 30-day and 90-day paid retention, net revenue retention on usage, gross margin trend, and ARR per employee. Pipeline metrics describe very little here.

Shooting Star

Bessemer’s Shooting Stars scale more like exceptional SaaS companies: strong product-market fit, customer retention and expansion, roughly 60% gross margin, and a Q2T3 path from about $3M in year one to $103M in year four. Many examples here are enterprise sales-led, but the sales motion is not the definition.

Where it breaks. Sales capacity. If revenue can only grow by adding reps, the curve is linear no matter how good the product is. The Shooting Stars that accelerated all had expansion inside the account doing part of the work.

Metrics that follow. Net revenue retention, pipeline coverage by segment, sales efficiency, time from pilot to production deployment, and reference velocity.

Compounder

Doubling annually or better, often cash-flow positive early, with 18 to 36 months for the $1M-to-$20M leg. Most good software companies live here.

Where it breaks. Usually nowhere. It breaks as a career problem when a board that funded a Compounder starts quoting Supernova numbers.

Metrics that follow. Net revenue retention, payback period, gross margin, and growth efficiency. These are the metrics a Compounder should be arguing for when handed a Supernova target.

Cautionary Tale

Real early demand with no moat, no margin, no retention, or a number nobody can defend.

For roughly a year, a Cautionary Tale can be indistinguishable from a Supernova. The warning category includes early demand accompanied by weak, disputed, or undisclosed evidence of durability. Jasper reached $42.5M in year one and peaked at $75M. Bolt.new posted the fastest $0-to-$20M on record. Both looked like category winners at month 12.

What separated them, visible only later. Jasper’s capabilities were absorbed into the frontier models’ free tiers. Bolt’s growth was not supported by retention or margin. 11x reported $14M contracted while former employees put surviving revenue near $3M. Cluely’s CEO built real distribution and then admitted the $7M ARR he had promoted was fabricated.

The tells at month 12, when the curve still looks identical: gross margin that is negative and unexplained, cohort retention that nobody will show you, an ARR figure whose basis changes between tellings, and a product whose core capability is a plausible feature of the platform it sits on.

Leaving the list

Companies do move off this list, in two different ways.

Higgsfield outgrew it. Forbes investigated the company in February 2026 at $300M. By August 2026 it had raised $400M at a $5.4B valuation on $700M in annualized revenue, quadrupling in eight months, with infrastructure costs down 45% after a cloud migration. It is classified as a Supernova in the dataset. What has not changed is that it discloses neither gross margin nor retention, in the most compute-intensive category in AI, so the two conditions that decide whether speed lasts are still open.

Jasper changed archetype. Its capability was absorbed, its founders stepped down in September 2023, and a new CEO moved the company from prosumer to enterprise and then into AI-search visibility, a workflow the platforms do not own. Jasper now reports enterprise ARR tripling year over year and more than 900 enterprise customers. But no independent total ARR figure has been published since 2022, enterprise logos moved only from 850 to 900 across seven months, and customer count fell from 120,000 to roughly 100,000 between late 2022 and early 2024. Segment growth published without a denominator is the same tell listed above.

The pattern worth taking from both: a rescue is a change of archetype, not a recovery of the original curve. It took Jasper a new CEO and three years, and what recovered was a segment on a Shooting Star path, not the Supernova curve the company started on.

The four Ps, by growth curve

The fundamentals did not change in the AI era. What changed is which of them carries the most weight, and how early the work has to be done. Same four questions, four different sets of answers.

Product

Supernova: The product is the distribution. The question is what leaves the building when someone uses it, and whether the free tier is generous enough to produce that artifact.
Shooting Star: The product must be referenceable. Security posture, compliance, deployment time, and measurable outcomes are product attributes that marketing depends on.
Compounder: Depth and reliability, plus a second and third product to expand into.
Cautionary Tale: The product is a feature of something larger, and the roadmap does not have an answer for that.

Price

Supernova: A credit or usage ladder can let an account grow 10x with no conversation. Replit’s inflection coincided with a move from flat-rate to usage billing and a reported gross-margin improvement from -14% to +23% in a year. Pricing is a high-leverage cross-functional decision on this growth curve, and marketing should bring customer evidence into it.
Shooting Star: The first seven-figure contract sets the reference price for every deal after it. Land-and-expand structure matters more than the entry number. Synthesia reports net revenue retention above 140%.
Compounder: Tiering and packaging discipline, reviewed on a cycle.
Cautionary Tale: “Unlimited” plans and subsidized inference, sold below cost to win share, then throttled when the bill arrives.

Place

Supernova: The loop is the channel. Community, creators, and the shared artifact. Gamma put 70% of its creator budget into micro-creators on performance terms.
Shooting Star: Design partners, alliances, and ecosystems. Harvey’s PwC alliance was distribution, not a press release.
Compounder: A mix, usually a self-serve floor plus partners plus a sales team, each carrying a known share.
Cautionary Tale: Rented distribution. It stops the day the payments stop.

Promotion

Supernova: The founder is the primary channel in year one. Proof artifacts do the rest. Lovable’s first year was founder posts, community, and a reseller program.
Shooting Star: Published outcomes, marquee logos, analyst validation, and a name for what the buyer is doing.
Compounder: Brand and content built over years, which is exactly why it compounds.
Cautionary Tale: The number is the marketing.

Diagnosing your own growth curve

Four questions, answerable in an afternoon with help from Finance and Product.

Can revenue grow faster than seller headcount? If usage, expansion, deal size, an ecosystem, or an installed base can carry it, a Supernova growth target is at least arithmetically possible. If nothing can, it isn’t.
What is gross margin, and where is it heading? Margin alone does not determine the archetype. Roughly 25% and improving can support Supernova economics; roughly 60% is closer to Bessemer’s Shooting Star benchmark; negative and unexplained is a warning sign.
What does the ARR number actually measure? Run-rate, contracted, consumption, or gross. Ask before you publish it.
What does the board think it funded? If nobody can answer this in one sentence, the metrics are already wrong.

The condition-by-company matrix, 26 companies across all four archetypes →