Whoever ships the agent that does the work sets that price. You can, if you pick the unit, put an agent behind it inside your product and price it as a line beside the seat before an entrant prices it for your customers.

• 4

  Funding rounds for AI-native entrants that run on existing systems, announced 14 to 28 October 2026.

  Source: FreightWaves; Tech Funding News; The Next Web
• $167M

  Raised to date by the four, as announced: $9M, $48M, $54M and $56M. Evos's figure is its pre-seed.

  Source: each company's October 2026 coverage, listed below
• 1 of 4

  States its price: Fragment charges on the amount of work its agents do.

  Source: Tech Funding News, 28 Oct 2026

What did these four startups announce in October?

Evos, Duvo, Hyphenate and Fragment announced funding between 14 and 28 October 2026, and each describes its agents as working inside systems the customer already runs. Rounds are given exactly as each company announced them.

• Evos raised a $9M pre-seed, announced on 14 October and reported by FreightWaves on 16 October. Its AI operators learn how employees make decisions and work inside the company's existing systems. At MyFreightWorld, a freight brokerage, one audits less-than-truckload carrier invoices against the TMS.
• Duvo raised a $32M Series A, $48M in total (Tech Funding News, 14 October). Its agents carry out approved steps across ERPs, supplier portals, spreadsheets and inboxes, through APIs where they exist and through the user interface where they do not.
• Hyphenate, formerly Maximor, raised a $45M Series A, $54M in total, according to a contributed article in The Next Web on 27 October. It layers onto the ERP a finance team already uses: The Next Web cites SAP, and HackerNoon (1 October) cites NetSuite and Intacct.
• Fragment came out of stealth on 28 October with $56M across its seed and Series A (Ventureburn, 29 October; Tech Funding News reports the round as a Series A). Its agents resolve direct-procurement exceptions end to end, with no data migration.

Four entrants, side by side

As announced in each company's October 2026 coverage. Outcome figures are omitted here because each company measures a different thing.

Raised to date by four AI-native entrants, as announced, October 2026

US$ millions, scale 0 to 60

• FragmentSeed plus Series A
  $56M
• HyphenateTotal, incl. $45M Series A
  $54M
• DuvoTotal, incl. $32M Series A
  $48M
• EvosPre-seed
  $9M
• 0204060

What exactly are they selling to my customers?

They sell units of finished work that a CFO can count: an audited carrier invoice, a reconciled supplier bonus, a resolved procurement exception, a closed month. Evos says its operator made about 3,500 audit decisions in two months at MyFreightWorld (company-reported, via FreightWaves, 16 October 2026). Duvo says Notino's quarterly reconciliation of about $45M in annual supplier bonuses fell from 70 days to 15 (company-reported, via Tech Funding News, 14 October 2026).

Each entrant also describes how it learns the judgement behind that work. Urav Shah, Evos's chief executive, told FreightWaves that in many of these businesses the most capable system is "the person who has been there 20 years", and Evos acts on its own once its decisions consistently match experienced staff. Duvo maps workflows through AI-led interviews, screen recordings and document analysis. Fragment's Autonomous Context Engine captures the policies staff actually use. Hyphenate trains on historical spreadsheets, emails and system logs and reuses resolved exceptions in later cycles.

Our reading: every resolved case improves the next one, and the improvement accrues to whoever recorded the resolution. Where an entrant does the work, that record sits with the entrant.

Your customers keep paying for the system of record, and a second vendor now bids for the work done inside it.

What does pricing by the unit of work do to my seat licence?

A price per unit of work charges for the finished job, so it competes with your seat licence for the same staff hours. Fragment is the one entrant of the four that has described its pricing. It charges "based on the amount of work their agents do" instead of by user seat or headcount, Tech Funding News reported on 28 October. Duvo and Hyphenate have not disclosed pricing. Duvo's founder, Tomáš Čupr, said in Tech Funding News's 14 October report that headcount is the wrong unit for measuring AI.

The hours are where this reaches your licence. Duvo says the specialist hours in each Notino reconciliation cycle fell from about 120 to under 10 (company-reported, Tech Funding News, 14 October 2026). Our reasoning: when the hours spent inside a system fall that far, the seats priced on those hours are what the next renewal tests. A seat priced on a specialist's month assumes the specialist still spends that month in your screens.

Who is paid for the work

The entrant grows by adding units of work. Hyphenate reports that 40% of its customers expand in their first year and that the average customer runs six modules (company-reported, in The Next Web's contributed article of 27 October 2026).

For a vertical SaaS company the open question is who sells that unit to your customer. The work happens in your product, on your data model, behind your approval screens.

What do the entrants still need from your product?

By their own coverage, the entrants still need new workflows to keep growing and the trust of the staff who sign off on the work, and both sit close to your product. Tech Funding News made the point about Fragment itself on 28 October: if its agents work, exceptions fall, so it needs new workflows to keep growing. RuntimeWire on 16 October named Duvo's next test as repeat deployments without a services-heavy model.

Trust is the other limit. In a survey Hyphenate commissioned, 14% of CFOs said they completely trust AI to produce accurate accounting data on its own and 97% said human oversight is critical (company-commissioned, reported in The Next Web on 27 October 2026). The person checking it signs off in an approval screen, and in your category that screen is in your product.

The vendor of record starts from a different position. It already holds every workflow in its category, the approval screens where staff sign off, years of each customer's records in the system the agents read from, and the customer relationship that renews each year.

What should a vertical SaaS leader ship in response?

1. Pick the unit. Choose the one unit of work in your category that an entrant would bill for, such as the audited invoice, the resolved exception or the reconciled period, and write down the test that says a result is correct.
2. Ship the agent that produces it. Build it inside your product, under your brand, and record the reason behind each correction your customers' staff make, so the next case starts from that reason.
3. Price the unit. Add it as a line beside the seat and take it into your next pricing review with the hours evidence from the exercise below.

The test comes first because a per-unit price needs an agreed definition of done, or every invoice becomes a dispute. The recorded corrections come second because they are what lets the agent improve on each account, which is the same advantage the entrants are building from outside.

The other threat, a customer replacing your product outright, is the subject of how to measure AI feature adoption in SaaS.

Where SynOS fits

SynOS runs inside your product, installed in your cloud and white-labelled, so the agent that audits the invoice or resolves the exception ships under your name and your customers see only you. It builds a Brain for each customer from that customer's data and procedures, traces and scores every agent run with writes recorded rather than executed, and reports cost per run by model and provider, so your pricing team has a record of each unit of work before it sets a price. Your first workflow can be live in your product in weeks one to two.

Questions