AI Product Updates, August 11: Copilot Gets a Cost Ledger, Manus Schedules a Data Reset, and Agent Workspaces Multiply (2026)

The biggest operational change on August 11 was not a new frontier model. GitHub made Copilot spend easier to explain, Manus announced a dated data-migration window that can interrupt affected accounts, and five same-day launches put more machinery around coding, video, and agent analytics.

Coverage window: August 11, 2026, 00:00–23:59 (UTC+08). The Product Hunt entries below are launch pages marked "Launching today"; their capability and performance descriptions are claims from the makers, not independent benchmarks.

Quick scan

Product	Releasing entity	Release date	What changed	Availability
GitHub Copilot usage report	GitHub	Aug 11	Adds per-model input, output, cache-read, and cache-write token details beside AI credits	Available to Copilot Business and Enterprise admins, and Copilot individual users 1
Manus account data transition	Manus	Aug 11	Certain data created on or after Dec 29, 2025 will be deleted Aug 23–24; backup and restore windows are provided	Affected users: backup before Aug 23, restore from Aug 25; unaffected users continue normally 2
Xirp	Spotify	Aug 11	Agentic development environment with institutional memory linked to services, ownership, docs, and architecture decisions	Free launch on Product Hunt 3
Bullet	Bullet team	Aug 11	Coding agent that selects model/reasoning level and parallelizes search, reads, and commands	Free launch; works with Claude Code or Codex subscriptions, API keys, or an on-device model 4
bb	bb team	Aug 11	Provider-neutral agentic workspace that can extend itself through prompts	Free launch; supports Claude Code, Codex, OpenCode, and more 5
Vizard Agent	Vizard	Aug 11	One agent for editing, generation, repurposing, localization, and revisions across a video job	Free options; launch page available today 6
Kubit	Kubit	Aug 11	Connects agent traces to user behavior and sends findings into coding workflows	Free options; new "Product Analytics for Agents and Users" launch 7

GitHub Copilot: AI credits get a cost ledger

Product: GitHub Copilot usage report Entity: GitHub Release date: August 11, 2026 Availability: Copilot Business and Enterprise admins, plus Copilot for individuals

The usage report now shows, for each model, the input, output, cache-read, and cache-write tokens behind the AI credits it consumed. Before this change, the report showed the credit total without the token anatomy underneath it. 1

That turns a monthly bill from a single number into something an engineering or finance team can investigate. A rise in input tokens points to prompt or context growth; cache-read and cache-write figures help separate repeated context from fresh work. Those are not new controls, but they make the existing usage easier to attribute.

The report is downloaded from the AI usage page in billing settings. For teams with chargeback or budget reviews, the immediate move is simple: save a baseline by model before changing prompts, routing, or agent settings. Otherwise a lower bill may just mean less work, while a higher bill may reflect more cacheable context rather than more useful output.

Manus: the important date is August 23

Product: Manus account data transition Entity: Manus Release date: August 11, 2026 (the native page is labeled Tuesday, August 11) Availability: Affected users get a backup window now and a restore window from August 25; unaffected users need to do nothing

Manus says it will return to operating as an independent company. To comply with regulatory requirements in specific jurisdictions, data generated by certain users on or after December 29, 2025 will be deleted from 8:00 a.m. on August 23 through August 24, 2026 (SGT). Affected users can back up until 7:59 a.m. on August 23 and restore from 8:00 a.m. on August 25. 2

The notice says affected accounts will be unavailable during the two-day deletion period. Manus will notify affected users in the app and by email. People who registered with Apple ID or Facebook should check in-app notifications because Manus says it may not have their email address. The company also says it will not charge affected users during the August 11–23 backup period.

Manus explicitly says this is not the result of a data breach or security incident. The stated reason is the separation from Meta and regulatory compliance. That distinction matters: the operational response is still a data-preservation task, but it should not be described as a security incident.

If you use Manus, check the in-app notice first, then run the backup tool before the deadline. Backing up once is not enough if you create new tasks afterward; the notice says to back up again so the latest task data is included.

New tools: agents are becoming workspaces

The five launches below point in the same direction, but they solve different bottlenecks. Xirp and bb organize agent sessions and context. Bullet tries to reduce the time spent waiting for a coding agent. Vizard Agent wraps a video workflow in one agent. Kubit measures what happens after an agent is deployed.

Xirp: institutional memory for development sessions

Product: Xirp Entity: Spotify Release date: August 11, 2026 Availability: Free; Product Hunt launch page marked "Launching today"

Spotify describes Xirp as an agentic development environment with institutional memory. The listing says it connects sessions to services, ownership, documentation, and architectural decisions through Spotify Portal, so a new session starts with organizational context rather than a blank prompt. 3

The practical question is not whether memory sounds useful. It is whether the links stay current enough to prevent an agent from following yesterday's ownership map or an obsolete architecture decision. Teams considering it should test one real repository with known ownership changes and measure how often the retrieved context is correct before making it the default workspace.

Bullet: faster loops around familiar models

Product: Bullet Entity: Bullet team Release date: August 11, 2026 Availability: Free launch; supports Claude Code or Codex subscriptions, API keys, and an on-device model

Bullet says its coding agent auto-selects the model and reasoning level for each prompt, parallelizes searches, reads, and commands, and uses targeted code search instead of embedding the whole repository. It claims 95.8% on SWE-bench Verified and 119 seconds per task; those numbers are self-reported on the launch page. 4

The compatibility choice is the more consequential product detail. Users can bring a subscription, an API key, or a local model, so the tool is trying to optimize the loop around models developers already pay for rather than asking them to switch providers. Before adoption, compare the same repository tasks with the existing agent: time to first useful change, unnecessary tool calls, and the cost of retries matter more than the headline speed claim.

bb: a workspace that edits itself

Product: bb Entity: bb team Release date: August 11, 2026 Availability: Free launch; works with Claude Code, Codex, OpenCode, and other providers

bb is an agentic orchestrator GUI. Its distinctive claim is that the workspace itself can be changed by prompt: ask for a task tracker and it creates one, then creates a skill that teaches the agents how to use it. 5

That makes configuration part of the agent surface. It can remove the wait for a vendor to ship every workflow feature, but it also makes governance harder: a prompt can change the interface and the instructions used by later agents. A team should keep the generated skills under version control and review them like code, rather than treating workspace customization as harmless preference data.

Vizard Agent: one request for the whole video job

Product: Vizard Agent Entity: Vizard Release date: August 11, 2026 Availability: Free options; Product Hunt launch page marked "Launching today"

Vizard Agent accepts raw footage, an existing video, a URL, a script, an image, or only an idea. The listing says it can handle editing, generation, repurposing, localization, and revisions in one workflow. 6

The useful test is end-to-end, not whether it can produce one attractive clip. Give it a source video with a fixed brand vocabulary and ask for two languages, platform-specific cuts, and one revision. That exposes where the agent preserves timing, captions, rights-sensitive material, and version history.

Kubit: analytics for agent behavior

Product: Kubit Entity: Kubit Release date: August 11, 2026 Availability: Free options; new Product Analytics for Agents and Users launch

Kubit connects agent traces to user activity so product teams can inspect why users re-prompt, drop off, or convert. Its launch page lists integrations through OpenTelemetry, a customer-data platform, or a bring-your-own-warehouse setup, then routes those findings into a coding agent. 7

This addresses a gap in ordinary product analytics: an agent can take ten different paths to the same user goal, so a page-view funnel does not show where the interaction went wrong. Teams should pair trace data with user outcomes and token cost. A lower re-prompt rate is not automatically better if the agent also stops completing the task.

What to do today

Copilot administrators: download the new usage report, save a per-model baseline, and separate input, output, and cache activity before changing budgets.
Manus users: check whether the account is affected, back up before 7:59 a.m. SGT on August 23, and repeat the backup after creating new task data.
Engineering teams: evaluate Xirp, Bullet, and bb on the same repository tasks; track latency, context accuracy, tool-call waste, and governance overhead.
Creative teams: test Vizard Agent on a real source video with localization and one revision, not just a single polished demo.
AI product teams: use Kubit-style trace and outcome data together. Agent quality is the completed user task, not the number of successful model calls.

The day's releases are a reminder that model access is only one layer of an AI product. The operational layer around it—cost visibility, retained context, workspace control, media workflow, and user-outcome measurement—is where many teams will feel the next product changes first.

References

1
Per-model token breakdown in the usage report
github.blog
2
A Note to Our Users
manus.im
3
Xirp on Product Hunt
producthunt.com
4
Bullet on Product Hunt
producthunt.com
5
bb on Product Hunt
producthunt.com
6
Vizard Agent on Product Hunt
producthunt.com
7
Kubit on Product Hunt
producthunt.com