Ballet's bet: no-code automation failed on review, not on authoring

Daniel Kimber and Ajain Vivek, the founders of Brainfish, have opened early access to a second product, Ballet, which takes a plain-English description of an outcome and generates reviewable, version-controlled integration code against internal databases and third-party software ^[1][2]. The notable design decision is not the authoring step, which most automation vendors now offer in some conversational form, but the artifact left behind afterwards.

The implicit diagnosis is that visual builders were never hard to author. They were hard to review, diff, test and hand to somebody else two years later. Ballet's response is to produce code an engineer can inspect and approve, and to split a workflow into deterministic steps, where the same input should produce the same result, and agentic steps that require judgment ^[6][7]. That distinction matters in proportion to blast radius: technical review gets more consequential when the automation can edit a live CRM record, change a price or push data into a billing system ^[20].

The target queue is specific. Ballet's early-access program is aimed at RevOps, Support Ops and product-led growth teams that know the outcome they want and the data involved but lack the engineering capacity to build and maintain the integration ^[5], the same teams whose tickets sit in backlogs ^[1]. For Brainfish, founded in 2022 to organize product knowledge so AI support agents could answer customer questions, this is a move from answers to actions: routing a lead, detecting account surges, changing an order, triggering a renewal play ^[3][4].

The published examples are the sort of plumbing that usually loses the prioritisation fight: enriching and deduplicating inbound leads before they reach Salesforce, joining product telemetry with Clay and ZoomInfo to alert sales reps in Salesforce and Slack, and propagating pricing or SKU changes from Salesforce CPQ into Stripe, an internal billing service and Freshdesk ^[8]. Those are Ballet's proposed use cases rather than documented customer deployments, and its claims about consistency, auditability and maintenance have not been independently benchmarked ^[8][9]. The current motion is consultative: a 30-minute session on the customer's real systems, then code for the customer's engineers to review ^[11].

The competitive field is not empty. Make's Maia builds scenarios through natural-language conversation, Zapier's MCP exposes more than 9,000 app connections and over 30,000 actions to AI tools according to Zapier, Pipedream mixes prebuilt actions with Node.js, Python, Go and Bash, and n8n combines visual workflows, code and AI components ^{[12][13][14][15]}. Zapier's own numbers work out to roughly 3.3 actions per connection, which is breadth, not depth ^[16]. Ballet is wagering that breadth is not the binding constraint on the workflows that stay stuck.

The founders' history is relevant to the narrow focus. Kimber studied Information Systems at the University of New South Wales while working at SiteMinder, where Forbes Australia reported he led the migration of 36,000 customers to a modular platform, and later held product roles at Willow and me&u; Vivek, Brainfish's CTO, has worked across architecture, infrastructure and automation ^[17][18][19]. A Brainfish account of an earlier failed venture says the pair went six months without speaking to customers and misunderstood the buyer before resetting at the end of 2022 [c20a].

Three things to watch. Whether design-partner engineers actually review the generated code or rubber-stamp it, in which case the audit story is theatre. Whether the code survives amendment by hand without regeneration destroying it. And what Ballet's own timeline means: its signup page previously said "Coming June 2026," and the available material does not establish whether that was a launch, an early-access opening or a target that has already slipped ^[10].