Szept

Szept costs $9.99 one-time for a personal Mac license and $199 for ten named seats, which means a dev-focused agency can equip an entire delivery pod for less than one billable hour. The strategic stake is not dictation speed; it is that voice, annotated screenshots, and screen recordings all land in Claude Code or Cursor as testable context, so senior engineers stop burning retainer hours on context transfer. Agencies without Mac-based technical clients should ignore it entirely.

Szept only pays off for agencies whose clients run Mac-based product, engineering, or design teams. The Personal plan costs $9.99 one-time, so a solo dev retainer can absorb it without a line item. The Team plan at $199 covers ten named people in one company with MDM deployment, which suits an agency standardizing dictation and agent handoff across a client's engineering pod. The workflow itself is the test: hold fn to dictate at the cursor, Command-Shift-A to annotate a screenshot, then paste testable context into Claude Code or Cursor. If a client's developers already live in those agents, the setup time is measured in hours. If the client is Windows-based or non-technical, there is no fit at any price. Run the matrix before quoting the retainer, not after.

Scaffold Versus Stewardship Split separates two jobs agencies routinely blur: generating net-new code fast, and keeping existing code correct as dependencies move. Scaffolding rewards speed, so agentic builders such as Verdent, which produces full-stack apps from plain English, compress a two-week prototype into days. Stewardship rewards coverage and traceability, so maintenance agents like Ripple, which scans repositories and opens pull requests when an API contract breaks, earn their value on the second and third client engagement, not the first. The split matters because the two jobs carry different failure modes: a weak scaffold is cheap to discard, while a missed breaking change reaches production and lands on the retainer. Price and staff them separately. Forrester's November 2026 research on vendors controlling metering, feature access, and dispute resolution shows why stewardship work needs an exit path, since the tool holding your client's maintenance history also sets the terms.

Agent Surface Area is the count of files, contracts, and downstream consumers a single AI coding run can touch before a human reads the output. It is the variable that separates a cheap scaffolding win from an expensive delivery incident. A prompt that generates a new internal tool touches one repo; a prompt that regenerates a shared API contract touches every client integration that imports it. Agencies should size each delegation by blast radius, not by prompt length. Ripple exists precisely because contract changes ripple outward: it detects a breaking change to an OpenAPI, Protobuf, or GraphQL spec and opens pull requests across every affected repository. Pair that with visibility tooling such as Agent Monitor, which graphs files and functions an agent is editing in real time and flags per-file risk before commit, and the surface becomes measurable. Cap autonomy where surface area is wide, and reserve full agent runs for greenfield work with no consumers.

Buy the $199 Team tier for up to ten named people when the whole delivery bench is on Mac and agent-assisted coding is billable, and skip Szept entirely for Windows or mixed-OS teams.

Route AI code tools to scaffolding, boilerplate, and cross-repo maintenance work, and keep architectural decisions, data modeling, and release sign-off with a named human owner.

IF your delivery team works on Macs and already runs Claude Code or Cursor daily, THEN the $9.99 one-time Personal licence (12 months of updates, then renew at $9.99 a year or stop) is a trivial test, and the $199 Team tier covers ten named people on one invoice with an MDM deployment kit. IF your client work is Windows-based, or you need a white-label surface to resell, THEN skip: Szept is Mac-only and there is no verified white-label program, so client-facing surfaces display the vendor's branding.

A five to seven day engagement that installs Szept on a client's Mac fleet, wires the Soniox speech key and DeepSeek cleanup endpoint, and trains developers to dictate, annotate, and hand off context to Claude Code or Cursor.