In partnership with

ShanghAI - Inside Doubao Work: ByteDance's $29B Bet on Your Business, Not Your Chatbot Habit

Hey Lobsters, welcome to this new edition of ShanghAI.

Here's the spreadsheet that forced ByteDance's hand: Doubao, its consumer chatbot with 200M+ daily users, generates under Rmb 1 million ($29 billion) — roughly 60% of last year's entire profit pool.

No amount of free-tier chat queries services that bill. So ByteDance did the Anthropic math: ~30 million daily users and enterprise-anchored revenue (Claude Code, API consumption) got Anthropic to a $965 billion valuation, past OpenAI.

ByteDance execs visited Anthropic this spring. The org chart moved fast after that.

The timeline is now confirmed, not speculative: Feishu (Lark) was folded into Doubao on July 30, 2026. Trae and Coze followed in late August, both now reporting to Doubao head Zhao Qi.

Doubao Work — the unified brand — went live August 29, 2026. This isn't a roadmap slide anymore. It's shipped.

For operators, the number that should reroute your Q4 budget isn't the $29B — it's Volcano Engine's rate card. Doubao-Seed-2.1 Pro runs ¥3.2/1M input tokens (~$0.44/1M) at 0–32k context, and prompt caching cuts recurring input costs to 20% of standard price.

Compare that to what you're paying per seat for Copilot, Zapier, a standalone agent builder, and Notion/Slack — roughly $130–150/mo stacked per user — and the arbitrage isn't subtle.

OUR PARTNER

The New Rules of Online Visibility

Your customers are searching in places your strategy doesn’t reach.

So before your business is buried and left behind, you need to understand the new rules of SEO.

BELAY's SEO in the Age of AI report explains how search is changing, what AI means for your visibility, and the practical steps small businesses like yours can take to stay visible.

BELAY’s U.S.-based Marketing Assistants turn strategy into execution, helping your business stay visible, credible, and competitive in every search.

HOW IT WORKS
REPLACE A ~$130–150/MO PER-SEAT DEV-TO-OPS STACK WITH ONE NATIVE EXECUTION LOOP, PRICED PER TOKEN INSTEAD OF PER SEAT.

The legacy stack doesn't fail inside any one tool — it fails at the seams.

A Copilot suggestion, a Zapier trigger, and a Notion update are three systems held together by webhooks, each hop adding latency, retry logic, and an auth token that can expire mid-workflow.

Doubao Work collapses that into one workspace:

  1. A Coze agent watches a trigger — a failing test, a support ticket spike, a stale doc — through native connectors inside the same workspace, not a third-party iPaaS layer.

  2. It acts directly inside Trae — opening a branch, running a build — because coding and agent orchestration now share one execution environment.

  3. It writes the result into Lark for human sign-off, in the same thread the work started in.

The unit economics: Doubao-Seed-2.1 Pro sits at roughly $0.44/1M input tokens (scaling with larger context windows), coming in at a fraction of Western frontier API costs.

If you are running high-frequency agent loops with static system prompts, context caching slashes recurring input costs by 80% after the first call.

For background triage and routing where you don't need heavy reasoning, Doubao-Seed Mini starts at under $0.03/1M input—cheap enough to process every inbound message without ever worrying about the compute bill.

SEE ALSO
The new tool of Jira

AI can build faster. Can your team decide better?

AI can draft the PRD and prototype the idea. Jira Product Discovery helps teams decide whether it belongs on the roadmap. Bring feedback and ideas together, prioritize as a team, and keep your roadmap connected to delivery in Jira.

WHAT TO USE
TOP 3 Models for Enterprise Workspace Automation (September 2026)

Simple flat illustration of a podium of a top 3 best AI models with a question mark for each position
  • BEST QUALITY/PRICE: Doubao-Seed-2.1 Pro/Turbo (Volcano Engine) ¥3.2/1M input (~$0.44/1M), strongest for long-context structured extraction and internal ops routing.

  • BEST LOCAL: Qwen2.5-Coder (0.5B–72B params, 128k context) — open-weight, Apache-licensed across most sizes, fully on-prem deployable for zero-data-leakage repos.

  • ShanghAI CHOICE: Doubao-Seed-2.0-Code + Coze-native agents — context caching cuts recurring input costs to 20% of standard price, making high-frequency agent loops close to free at scale, without the webhook layer where most SMB automations actually break.

  • THIS WEEK OFFER (Affiliate): Notion Business - 3 MONTHS FREE

APPLY TO YOUR BUSINESS
HEADLESS AD FACTORY: COZE → SEEDANCE 1.5 PRO → LARK APPROVAL → AD PLATFORM PUSH

This architecture replaces an external creative agency, a media buyer, and a landing page developer with a single continuous pipeline running on native ByteDance infrastructure.

The process begins inside Coze Enterprise, which continuously monitors live sales spikes from your Stripe or Shopify API.

When a specific product variant crosses a 15% daily conversion threshold, Coze automatically generates a creative brief containing existing product assets, ICP parameters, and validated ad copy angles.

This brief routes directly to Seedance 1.5 Pro via Volcano Engine API, generating four distinct UGC-style video variations with alternate hooks in under 90 seconds.

Prompt caching keeps the recurring brand-guideline context costs below a cent per generation.

Once assets are rendered, an autonomous sub-agent inside Trae generates a dedicated, localized landing page branch with live stock counters and conversion-optimized checkouts.

The entire package—video renders, ad copy variations, and preview URL—is pushed into a single Lark Base thread.

A growth lead reviews the creative on mobile and taps an approval button. That single tap triggers an automated webhook call to the TikTok and Meta Ads APIs to launch the test campaigns with predefined budgets and tracking parameters.

Setting this up shrinks asset turnaround from five business days to six minutes, dropping creation costs from $200 per video to roughly $0.80 in total compute.

Implementation & Enterprise Advisory:

If your engineering team is evaluating Doubao Work, private DeepSeek clusters, or open-weight inference stacks, ShanghAI audits architecture, optimizes token spend, and delivers enterprise-grade deployments with zero US vendor lock-in. Book an Enterprise Stack Audit with LeanAI.

The AI IPO Rush Is Coming

OpenAI and Anthropic could bring a new wave of AI attention to the public markets. But investors don’t have to wait for the IPOs.

MarketBeat’s 7 AI Stocks to Buy Now report reveals 7 publicly traded companies positioned to benefit from the next phase of AI investment.

Until next week,
The ShanghAI Guy