Skip to content

Sprint - October 19 to October 30 #116781

Open
Open
Sprint - October 19 to October 30#116781
Labels
sprintSprint planning

Description

@langesven
langesven
opened on Oct 14, 2026

Sprint Planning

List of incidents + any action items that haven't been scheduled yet

Team sprint planning

For your team sprint planning copy this template into a comment below for each team.

# Team ___

**Support hero:** ___
**Off during the sprint:** ___

## Quarter goals

1. Quarter goals, status (red/yellow/green) and action points if yellow/red

## Retro

### Tasks

<!-- Grab the high and low priority items from last time and add whether that item was completed or not -->

## Plan

### High priority

-

### Low priority / side quests

-

### Are any other teams impacted by this plan? If so, tag them here

-

Activity

  1. langesven
    added
    sprintSprint planning
    on Oct 14, 2026
  2. langesven
    pinned this issue on Oct 14, 2026

  3. langesven commented on Oct 14, 2026

    @langesven
    langesven
    on Oct 14, 2026
    ContributorAuthor

    Team Cloud Platform

    Quarter Goals

    Retro

    Details

    High Prio

    • ✅ Platform UI/API service catalog PoC
      • ✅ A one-pager service catalog with service health, promo state, curr freight for Kargo + argo @k33lan
      • ✅ A thin API returning above + what stage a PR is in (where is my change?) @k33lan
    • Agents
      • ⚪ Expose deploy-freeze workflow via Sherlock MCP @k33lan
      • 🟡 Close investigation quality gap Sherlock <> PostHog @k33lan
    • 🟡 More aurora_user_management/v3 work, pgbouncer hubs (compute), cross-az reduction (traffic) @langesven
    • ✅ Deployments - implement easy/golden path for helm charts that wrap infra we have (non-posthog app) @Z3r0Sum
    • 🟡 Deployments - more migrations @Z3r0Sum (perpetual 😄 ).

    Low Prio

    • ✅ Q4 Tracking issue for Unified Platform goal @k33lan
    • Migrate PostHog Django stack to platformctl @langesven
    • 🟡 General research for Q4 goal as far as available tooling goes @langesven
      • Set up Kviklet as a PoC for possible audited prod db access
      • Look into Metabase featureset again, maybe it's useful for write access?
    • ✅ Deployments - platformctl native service repo support (i.e., the hogtower use case) @Z3r0Sum
    • 🟡 Deployments - enable / get working some of the instrumentation for services like kargo-mcp / deployment-events @Z3r0Sum

    Side Quests

    ✅=finished 🟡=in progress 🔴=won't finish ⚪=not started

    Plan

    High Prio

    • per-AZ PgBouncer Hub setup, take it live in prod @langesven
    • per-AZ deployment feature for posthog-app? @langesven
      • capture and others are VERY sensitive to uneven scaling since PreferClose was introduced
      • best way to deal with that might be making the per-AZ PgBouncer logic a thing we can also apply to Deployments (contour-style, but cleaner)
    • automated setup of new Postgres DBs into Metabase @langesven
    • Hogstation
      • Hook up to PostHog for analytics pre-alpha rollout to engineers @k33lan
      • Stand up MCP v0 over API (could this be under SherlockMCP)? @k33lan
    • Agents
      • Expose deploy-freeze workflow via Sherlock MCP @k33lan
      • Can we close down Sherlock auto-triage? @k33lan
    • Deployments - streamline the migration process a little more so we can multiplex / focus on migrating more services @Z3r0Sum
    • Deployments - setup alerting / o11y for Kargo itself so our users aren't the first to report problems (maybe a hogstation dashboard for the platform team around deployment health too?) -> requires Kargo 1.12.0 upgrade @Z3r0Sum
    • Deployments/Hogstation - nice onboarding experience to Kargo rather than just pointing at docs @Z3r0Sum

    Low Prio

    • evaluate write-access feature in Metabase @langesven
      • can we do some sort of just-in-time/approval access?
      • can we run arbitrary SQL?
      • is it audited?
    • Hogstation
      • Nice domain name, hogstation/? @k33lan
    • Deployments - start to look into Kargo Sharding as this might be a bottleneck we need to solve before we're 100% onboarded @Z3r0Sum

    Side Quests

    • Cloud Platform Times! (This might make sense now & we can tie in Hogstation/Kargo/HostHog) @k33lan
    • Eli reported some issue with millions of stuck tasks in the Celery Redis in prod-us @langesven
      • thinking about setting up a new Redis in prod-us account and migrating jobs over, rather than trying to fix up the old root-account Redis
    • Hogstation - cool representation of a deployment flow ... hogs on a train? @Z3r0Sum

  4. MichaelKutsch-ph commented on Oct 14, 2026

    @MichaelKutsch-ph
    MichaelKutsch-ph
    on Oct 14, 2026
    Contributor

    Team Cloud Foundations

    Quarter Goals

    Mega issue: https://github.com/PostHog/posthog-cloud-infra/issues/10708

    Retro

    Details

    High Prio

    • ⚪ Vendor negotations @MichaelKutsch-ph @sinthetix - focusing on cost reduction first, negotiations on hold rn
    • 🟡 make managed proxy HIPAA compliant @MichaelKutsch-ph - trickier than we thought. currently working on cost allocation for the current proxy, then taking it from there
    • ✅ support ai-research with ML infra topics @MichaelKutsch-ph progress here, this will become an actual topic in q1, don't expect any major work rn
    • ⚪ smokescreen-sidecar canary w/ CDP @MichaelKutsch-ph - paused in favor of finops topics
    • 🟡 Implement CNPs / client SGs for MSK / WarpStream @Daesgar
    • 🔴 Move static assets upload from Argo to CI @Daesgar
      • Replaced (for now) with asset upload from web and frontdoor
    • ✅ Create safeguards / alerts for static assets @Daesgar
    • ✅ S3 / RDS cost analysis and tuning @Daesgar
    • 🟡 Network Cost report on production @danielxnj
    • 🟡 Managed proxy cost data @danielxnj - partially done? we have the cost data, but can't allocate cost per proxy rn
    • US EKS upgrade @danielxnj
    • ✅ grafana-alloy: scope the attach_metadata node watch to the local node @adickinson72
    • ✅ network-cost-exporter: scope the Node informer to the local node @adickinson72
    • ✅ otel-collector: drop the agent ReplicaSet watch @adickinson72
    • ✅ cleaned up kafka-deduplicator volumes @adickinson72
    • 🟡 Cilium endpoint health checking off @adickinson72
    • 🟡 tagging @sinthetix

    Low Prio

    • ⚪ contour upgrades @MichaelKutsch-ph
    • ✅ incident.io: page dev alerts that opt in via page_in_dev @adickinson72
    • 🟡 EKS module refresh @adickinson72 - own eks-posthog module replacing upstream v19; moves-only switch proven in playground

    side quest

    ✅=finished 🟡=in progress 🔴=won't finish ⚪=not started

    Plan

    High Prio

    Low Prio

    • ⚪ contour upgrades @MichaelKutsch-ph
    • add finOps skills so agents do the discovery for us instead of us poking once a year

    side quest

  5. tomasfarias commented on Oct 14, 2026

    @tomasfarias
    tomasfarias
    on Oct 14, 2026
    Contributor

    Team Batch Exports

    Support hero: @tomasfarias / @rossgray

    Quarter goals

    • Objective 1: 10% usage comes from HogQL-modeled batch exports
    • Objective 2: Understand and optimize our product's growth
    • Objective 3: Scaling batch exports to many more customers

    Retro

    Details

    ✅ Resource based allocator
    🟡 Auto-scaler for batch export workers

    • Still seeing out of memory errors.
    • Could still have memory leaks around.
      • May need profiling to find out where the issue is.
      • Network request build-up?
      • Try out different memory allocators for arrow.

    🟡 HogQL batch exports
    🟡 Support for different modes (only for warehouse destinations):
    🟡 APPEND
    🔴 FULL_REFRESH
    🟡 MERGE (requires both key fields)

    🔴 Deprecation of $set/$set_once

    • Applied it, realised would break http batch exports, rolled it back.
    • Needs HTTP batch export migration.
    • Discuss with other teams? How to avoid corrupting/missing data in a migration?
      • Could we use a different ingestion path (less writes)?

    ✅=finished 🟡=in progress 🔴=won't finish ⚪=not started

    High priority

    • HTTP batch export migrations

      • Enables removing data from events, leading to massive cost savings.
      • Work with ingestion to figure out a path forward for HTTP batch export migrations.

    • HogQL batch exports

      • Adding inserted_at columns (check with team CH).

    • Memory auto-scaler for batch export workers

      • Review memory usage, allocators.
      • Run tests.

    Low priority

    • Product isolation

    • Cost management

      • What's costing us more? Egress, workers, temporal, etc...?
      • Are there opportunities to optimize our costs?

    • Batch exports powers SQL editor

      • Decide billing (need RFC)
      • Ideally, keep this separate from the other work (different feature flag, A/B test)
      • Support more file formats

    • Batch exports powers insights exports

      • SQL insights first

    • Improve SLA alerts

      • Too noisy when continuous runs keep going over
      • We should look into switching to event based alerts, rather than log lines

    • Integration clean up

      • Redshift: migrate and clean up.
      • PostgreSQL: clean up.
      • Snowflake: clean up.

  6. meikelmosby commented on Oct 14, 2026

    @meikelmosby
    meikelmosby
    on Oct 14, 2026
    Collaborator

    Team Workflows

    Time Off

    Quarter goals - Q4 2026 objectives

    Details
    • 🟡 Be our own customer (@mayteio)
      • 🟡 Functionality gaps for team 2
      • ⚪ Start with single campaigns by hand
      • ⚪ Move rest over with importer
      • ⚪ Fully move over team 2
    • 🟡 Grow Workflows (@meikelmosby & @Silthus)
      • ⚪ Onboarding flow for Workflows (UI and MCP)
      • 🟡 Cross-selling & integration with other parts of PostHog
      • ⚪ Route existing marketing and messaging destinations into Workflows
      • 🟡 Behavioural cohorts
      • ⚪ Importers for competitors
      • ⚪ Engage with high intent customers to migrate them over or get them set up to use messaging
      • ⚪ Reach out to customers that use CDP marketing destinations
      • ⚪ Promotion campaigns
    • 🟡 Have self-driving v1 in GA (@dmarchuk)
      • 🟡 Proactively suggesting changes to a single workflow/broadcast from customer data
      • ⚪ Proactively suggesting changes across workflows/broadcasts from customer data
      • 🟡 Proactively suggesting new workflows/broadcasts from customer data

    ✅=finished, 🟡=in progress, 🔴=won't finish, ⚪=not started

    Retro

    Details

    @dmarchuk

    • Self-driving: test suggestions on team 2 and fix what comes up
    • Cross-selling & integration with other parts of PostHog
    • Node push registration service: dev rollout, then prod mirroring before switching traffic
    • Land destination secrets hardening
    • Broadcasts forward fixes and improvements (deleted sender on content step, add people from an uploaded list)

    Low priority:

    • Self-driving: prototype for suggesting new workflows (ideas scout, rotating cohorts of likely adopters, rejection reasons)

    @mayteio

    • Workflows product isolation to unlock CI skipping 🥳
    • Marketing Push for PostHog Tasks, Loops etc.
    • Marketing on workflows gaps
      • Team-wide frequency caps on emails
      • Per-recipient data from a CSV upload as liquid variables for broadcasts
      • Start a new email from a saved or pre-made template
      • Image library
      • Tagging system (taking over from @Silthus)
      • Index improvements / files?

    @meikelmosby

    • Finalised and production tested DLQ & replay
      • further harden VM and input safety due to wrong classification
    • Behavioural Cohorts (taken over by @mayteio)
    • fixed DNS and outbound request performance
    • workflows navigation revamp
    • workflows onboarding flow & wizards
    • disabled SES feedback forwarding & introduced configurable reply to (plus backfill)
    • workflow suggestions for onboarding (UI & MCP)
    • introduce workflows into the wizard

    This Sprint

    @dmarchuk

    • Self-driving: prototype for suggesting new workflows (ideas scout, rotating cohorts of likely adopters, rejection reasons)
    • help @mayteio to move team2 over to workflows

    @mayteio

    • Move PostHog to Workflows
      • Land behavioural cohorts
      • migrate transactional emails
      • Marketing migration goal (the most we can do)
      • UX improvements (tagging, start a new email from a template, Juliana's doc)
      • Have a tool / importer to migrate

    @meikelmosby

    • Nail Onboarding: introduce workflows into the wizard
    • Build Scheduling calendar & integrate it into onboarding and broadcast setup
    • customer interviews / shadows for onboarding
    • Route existing marketing and messaging destinations into Workflows
    • Hands-on help for teams moving to Workflows

    @abheek9

    • Handover document
    • Marketing pushes

  7. rorylshanks commented on Oct 14, 2026

    @rorylshanks
    rorylshanks
    on Oct 14, 2026
    Contributor

    Team ClickHouse

    AKA Team Read Fewer Bytes Please

    Quarter Goals

    🧬 Cell-based architecture @rorylshanks
    📦 JSON datatype in the app @rorylshanks
    🔐 ClickHouse is up to date and secure @tpgilmore
    🚚 ClickHouse migration system @orian
    🗑️ Data lifecycle management @orian
    🤖 Self-driving operations @bciaraldi

    Retro

    Details
    • Backfill all data, setup ingestion on test cluster for new schema @rorylshanks 🟢
    • Event deletion in the app @orian 🟡
    • Clean up secrets @tpgilmore 🟢
    • Dagster security cleanup @tpgilmore 🟢
    • Better AI Tooling in infrastructure / CH fork repos @juhop-code 🟡
    • Data lake architecture @juhop-code 🟢
    • APM ingestion migration @bciaraldi 🟡
    • Table moves off main @bciaraldi 🟢
    • Keeper upgrades / kernel upgrades / infra upgrades @bciaraldi 🟡

    High priority

    Low priority / side quests]

    🟢 =finished 🟡=in progress 🔴=won't finish ⚪=not started

  8. dustinbyrne commented on Oct 14, 2026

    @dustinbyrne
    dustinbyrne
    on Oct 14, 2026
    Contributor

    Team Client Libraries

    Support hero: @turnipdabeets then @dustinbyrne
    Next sprint planning @marandaneto
    OOO: 1 day @marandaneto, @ioannisj, @turnipdabeets

    Quarter goals

    Q4 2026 objectives

    • 🟡 Goal 1: Improve SDK observability and diagnostics (@marandaneto)
      • ⚪ Benchmark SDK performance
      • ⚪ Build dashboards to monitor SDK health and performance
      • ⚪ Capture diagnostic events to investigate SDK issues
      • 🟡 Add remote configuration support for SDK diagnostics
      • 🟡 Enable or disable diagnostics globally, per organization, or per project
    • 🟡 Goal 2: Expand SDK coverage and automate development with shared specifications (@dustinbyrne)
      • ⚪ Define canonical SDKs for frontend, backend, and mobile
      • ⚪ Run agentic loops for bi-directional sync of SDKs and specs, verified by a shared acceptance-test harness
      • 🟡 Expand first-class SDKs and integrations:
        • 🟡 Release @posthog/browser v2 GA
        • ⚪ Release @posthog/next v1 GA
        • ⚪ Release posthog-kmp v1
        • ⚪ Release posthog-android-surveys-compose v1
        • 🟡 Release the iOS SDK v4
        • ⚪ Add Laravel support
        • ⚪ Add Electron support
        • ⚪ Add Angular support
        • 🟡 Add Flutter desktop support through a confirmed community donation
          • 🟡 Review the donated implementation
          • ⚪ Test Flutter desktop support
          • ⚪ Release Flutter desktop support
    • 🟡 Goal 3: Bring mobile session replay to feature parity with web (@ioannisj)
      • ⚪ Decide whether to deprecate wireframe mode for mobile session replay, and proceed if agreed
      • 🟡 Add v2 session replay triggers to mobile SDKs
        • ⚪ Remote masking config
    • ⚪ Goal 4: Research how to capture product usage across agent-driven interactions (@turnipdabeets)
      • ⚪ Interview customers building agents, bots, MCP servers, and edge applications to understand what they need to measure, how they capture events today, and where our SDKs fall short
      • ⚪ Evaluate capture options, including deriving analytics events from tool calls, capturing agent-driven interactions through WebMCP, and offering lightweight SDKs for edge runtimes and unsupported ecosystems
      • ⚪ Share findings with the MCP and web teams, and recommend opportunities to test with customers
      • ⚪ Pair with the Product Analytics team to explore their ideas and test them together where client libraries support is needed

    Retro

    Details

    High priority

    • 🟢 Have a productive team offsite @PostHog/team-client-libraries
    • 🟢 iOS SDK next major version (on a 4.0 branch) @turnipdabeets @ioannisj
      • Remove already deprecated values
      • Move undocumented, built for wrappers, APIs into @_spi
      • Address SPI import in RN session-replay package
      • Raise the iOS minimum to 15
      • Remove Wireframe?
    • 🟢 Bridge iOS SDK gap with sdk specs @turnipdabeets
    • 🟢 v2 triggers for iOS, Android @ioannisj
    • 🟢 nextappcon prep + talk + booth work @marandaneto
    • 🔴 Capture diagnostic events to investigate SDK issues (python) @marandaneto
    • 🟢 rethink q4 goals (goal 4 and 5) after new direction @PostHog/team-client-libraries
    • 🟢 Release the @posthog/browser SDK extension changes @dustinbyrne
    • 🔴 @posthog/browser clean up, deduplication @dustinbyrne
    • 🟢 Expand Node.js SDK adapter support in the SDK test harness @dustinbyrne

    Low priority / side quests

    Plan

    High priority

    • Capture diagnostic events to investigate SDK issues @marandaneto
      • Also Benchmark SDK performance
    • 🟡 Add remote configuration support for SDK diagnostics @marandaneto
      • PR is open already, review and merge
    • 🟡 Enable or disable diagnostics globally, per organization, or per project @marandaneto
      • PR is open already, review and merge
    • 🟡 iOS major release @turnipdabeets
      • Flutter major
      • RN plugin major
    • v1 capture for iOS (part of major) @turnipdabeets
    • Prepare @posthog/browser documentation and pre-release rollout @dustinbyrne
      • migration guide
      • GA blocker list and rollout plan
        -@posthog/browser snippet installation @dustinbyrne
    • Land the iOS replay capture and masking rewrite @ioannisj
    • Replay remote masking config for mobile @ioannisj

    Low priority / side quests

    • Session Replay: SwiftUI extract text from accessibility tree PoC @ioannisj

    Are any other teams impacted by this plan? If so, tag them here

    🟢 =finished 🟡=in progress 🔴=won't finish ⚪=not started

  9. jurajmajerik commented on Oct 15, 2026

    @jurajmajerik
    jurajmajerik
    on Oct 15, 2026
    Contributor

    Team Experiments 🧪

    Retro

    Sprint #19, from Oct 5, 2026 to Oct 16, 2026

    Anders

    • Flags V2

      • Flag pages: v2 flags now show read-only in the flag list and detail pages instead of breaking them (#106774). Behind feature-flag-rules-v2-editor, boolean v2 flags can be created and edited in the UI, with ordered targeted-release and percentage-rollout rules (#107734).
      • Write errors: v2 write errors return the field path in attr and the bare message in detail (#109452). The editor shows each error under its field, and API and MCP clients no longer parse the message.
      • Raw readers: A repo test counts code that reads the flag filters document directly, and the semgrep rule catches more shapes (#109448). 75 sites remain, and the list can only shrink.

    • Refactors, fixes, etc.

      • Fingerprints: One calculation config builds the metric fingerprint at all 12 call sites (#109371). Experiments that count only matured users now launch with fingerprints that their daily results match.
      • Result store: All metric result reads (#109403) and writes (#109451) go through one store. A daily write no longer fails when a recalculation already stored a result for the same window. The "experiment completed" event and the concluded notification use one rule for the current result of the first primary metric (#109498).
      • Shared metrics: The API returns each shared metric with the experiment's overrides already applied, so the page and MCP no longer apply them on their own (#109357). The four breakdown edits share one save path and skip saves that change nothing (#113756).
      • Failed saves: A failed experiment save no longer shows success toasts, triggers a recalculation or captures date-change events (#113762). The CUPED, statistics and date dialogs stay open with the user's edit and show the server's value after a conflict (#113853).
      • Exposures: Experiments started before the 2026-09-01 cutoff read both $feature_flag_called and $experiment_exposure. They keep counting exposures after their organization moves to flag_evaluations (#114399).
      • Timeseries text for agents: The MCP timeseries text labels the real baseline instead of always control (#115227). It renders from a typed view, and the formatter moved into the experiments product (#115262).
      • Charts: Quill charts keep their last frame when their container collapses. This fixes the flaky multi-step funnel story in visual review (#113847).

    Juraj

    • Experiment query performance

      • Extended precompute to ratio metrics (#115871) and widened the auto-enrollment criteria so more experiments get it (#113405).
      • Tagged each precomputed read with a skip reason so we can see why a read fell back to the raw query (#113706).
      • Split the query runner into smaller pieces: moved out the precompute policy (#114688), the actors query (#115250), and the metric-events precompute (#116755).

    • Experiment fixes

      • Fixed the top errors from the weekly triage: scheduled recalcs crashed when the creator left the org (#116083), cohort filters pointed at deleted cohorts (#116790), and cloning an experiment failed on a taken flag key without saying which key (#117782).
      • Made filters like status >= 200 work on string properties (#111714) and let the API accept metric breakdowns (#113698).
      • Dropped the legacy recalculation time column now that two daily times are live (#115395).
      • Fixed the onboarding snippet for server frameworks (#112332) and unified text sizes in the experiment meta bar (#117674).
      • Audited the experiments docs: corrected wrong claims (#20721, #20778), refreshed stale copy (#20722), and added a Rails page (#20970).

    • Warehouse-native experiments

      • Added direct querying for BigQuery, no sync needed (#113994).
      • Added a warehouse source node and exposure config for experiments (#114634).
      • Wrote an RFC on whether warehouse metrics should be inline SQL or reusable metric sources (#1322).

    • Playground

      • Added tool messages: a tool role with structured tool calls (#112333), Gemini function call mapping in the proxy (#111709), and import of tool calls from traces (#113398).
      • Added move up and down controls for messages (#113963).
      • Fixed a provider name mismatch that broke completions (#113040).

    • Prompts

      • Made the SDKs share one request when many calls fetch the same prompt at once: JS (#5266) and Python (#1059).
      • Added viewing and unarchiving of archived prompts (#114048) and stopped the preview from showing stale results (#114040).
      • Removed the prompt partials flag (#111708) and wrote a guide for using prompts in Temporal workflows (#21180).

    Marcel

    • Diagnosis skill

      • #112315 - feat(experiments): refresh the experiment diagnosis skill and its evals
      • #112477 - feat(experiments): rename diagnosing-experiment-health skill to cover broader scope

    • Health layer scaffolding, beginning of migration & UI

      • #112916 - feat(experiments): add the experiment health layer scaffolding
      • #113150 - feat(experiments): port the flag-state and no-metric health checks to the backend
      • #113190 - feat(experiments): return health check findings on the experiment API
      • #113353 - feat(experiments): show the flag-state warning from the experiment's health findings
      • #110279 - feat(experiments): track health findings and every launch path
      • #113732 - feat(experiments): show health findings in one panel
      • #114109 - feat(experiments): run the exposure health checks on the server
      • #114297 - chore(experiments): tracking for health-check parity and launch findings
      • #114458 - feat(experiments): flag release conditions that force a variant
      • #114739 - refactor(experiments): one home for the health checks' flag rules
      • #115624 - feat(experiments): report a deleted flag on a running experiment
      • #115457 - feat(experiments): return health findings to every experiment reader (API only, no UI change)
      • #115784 - feat(experiments): report broken flag variants + time the health request

    • Other

      • Customer support
      • Replay vision team coordination for experiment scanner and desired features
      • #113739 - feat(experiments): search experiments by feature flag key
      • #111758 - feat(experiments): raise the freeze cap to 200k and show its duration

    Rodrigo

    • Metrics recalculation

      • Finished the switch off the timeseries publish pass: required a recent data window before skipping a scheduled run (#111679) and removed the publish pass itself (#111422).
      • Hardened recalculation under load: jittered the capacity retry delay (#113036), capped retries on capacity and timeout errors (#115794), kept precompute work inside the attempt budget (#115823), and killed precompute inserts when an attempt is cancelled (#115822).
      • Added a five-minute cooldown on manual reloads: block reloads after the latest run (#112207), return that latest run inside the window (#112205), and gate the rate limit behind a flag (#113949).
      • Made recalculation the single source of results: switched the notebook component to recalculation results (#114248) and removed the non-recalculations code path (#114305).
      • Tightened self-healing: heal the latest run only when a metric is missing (#115684) and skip known failed metrics on stopped experiments (#115796).
      • Sharpened the schedule-to-start metric buckets (#113613).

    • Loading-state fixes

      • Cleared metric loading indicators on network error (#112895) and skip latest-recalculation updates after unmount (#116340).

    • Experiments health

      • Built the experiments health category and its table (#115842, #115843), used the category table on the generic detail page (#115841), and added the job owners (#115835).
      • Added the health checks: flag state (#115837), forced variant (#115838), and no-metric (#115839), and mapped the findings to health issues (#115836).

    • Experiment header

      • Added a collapsible content slot to the scene title (#113645) and collapsed the experiment header behind the title toggle (#113646).

    • Product folder migration

      • Migrated the scene components to the product folder (#110387) and declared the frontend dependencies of the product package (#112194).

    Ruby

    • Wrote the Health Engine Needs research doc (research/2026-10-07-Health-Engine-Needs.md)

    • Roast changes to New Experiment UX: variant screenshots (#108053), Replay Vision card (#108107), step tracking (#108105), code as last step (#108106), auth-persistence question (#108082), wizard guide + Ask AI (#108052)

    • MCP Error reduction efforts: call-shape/alias telemetry (#101133), scalar constraints in exec info (#101164), redirect for removed experiment-get-all tool (#103254), collapse long prompt snippets (#102656).

    • Pricing/billing: grandfathered-plans RFC (#1240) updated

    • Prompts NPS launched Oct 1

    • NPS-to-Slack publishing bug in #team-workflows (still broken Oct 14).

    • PR #116867 - Papercuts: cohort add-users CTA reported in #papercuts; draft

    • ETag 304 billing impact sized (+$37k ≈ +11% US flag MRR); realtime-cohorts cost-tagging (charts #15897); Email comms to those impacted

    • Website investigation for product pages + use of scanner

    • Comparison of revenue attribution methods.

    • Supporting meetings on experiment health checks / mutually exclusive experiments architecture; ElevenLabs call on experiment isolation and multi-test interference.

    • Wrote up OST on new UX

    Sprint Goals

    Sprint #20, from Oct 19, 2026 to Oct 30, 2026

    Anders

    • Local end to end testing of flags v2 with SDK, local evaluation, etc

    • Spec proposal for API design for v2 changes ready for review

    • Remove the legacy query runners

    Juraj

    • Warehouse-native for single source MVP done end to end

    Marcel

    • Ongoing health layer work

    Rodrigo

    • Implement more complex, non-deterministic checks on the health workflow.

    • Improve performance of the daily scheduled recalculation to prevent capacity issues.

    • Some more tuning of the recalculation workflow.

    • Frontend refactor on the experiment metric logic to remove some redundancies.

    • Automate product migration.

    Ruby

    • Finish health engine research doc, and give feedback to team.

    • Write an RFC on no-code, incl. marketing ICP distribution across PH, and current experiment users. Incl. what companies are doing today, existing bugs raised, and whether our current approach is what we should be doing, until more investment.

    • Write up initial product page investigation into RFC concerning product pages for exp and feature flags on website, and improvements needed for website.

    • Consolidate how we validate products that we want to build - aka. eng delve in and how we can double check.

    • Investigate how we can build auto-flags cleanup

    • Make progress on cleaning up or tables

    • Warehouse-native needs: identifying users to share MVP with

  10. vdekrijger commented on Oct 15, 2026

    @vdekrijger
    vdekrijger
    on Oct 15, 2026
    Contributor

    Team Query Performance

    Support hero: @georgemunyoro

    ⚪ = Not Started · 🟡 = In Progress · 🟢 = Completed · 🔴 = Won't complete · 🔵 = Abandoned · 🟣 = Ongoing

    Quarter goals

    The Q4 goals and George's additional compiler objective are still open. These are the current proposals;

    • 🟡 Proactively improve customer experiences — Vasco. Customer-facing observability and resilience work is progressing (query latency phases #115652, dashboard correlation #115279); a measured end-to-end customer journey and proactive regression detection are not shipped yet.
    • Make query-performance experiments easy to run and trust — Andy.
    • Safely bring the JSON Events Table into production — Sandy.
    • Turn HogQL into an extensible optimizing compiler — George.

    Retro

    Notes: https://posthog.slack.com/docs/TSS5W8YQZ/F0B741LSUVA

    Tasks

    Andy

    • AL

    Sandy

    • AL

    Vasco

    George

    • AL

    Mike

    • TBD

    Plan

    High priority

    Sandy

    • TBD

    George

    • TBD

    Andy

    • TBD

    Mike

    • TBD

    Vasco

    • 🟡 Resilience / busy-hour ClickHouse congestion fixes to create headroom whilst work on the rate limiter and query router is on the way
    • 🟡 Continue with the performance / experience dashboards, time permitting
    • 🟡 Help Rory ship and validate the new ClickHouse fork / JSON events table

    Low priority / side quests

    Are any other teams impacted by this plan? If so, tag them here

    • ClickHouse (@rorylshanks): ongoing joint work on the new fork / JSON events table, production smoke tests, query correctness and safe rollout.

  11. lucasheriques commented on Oct 16, 2026

    @lucasheriques
    lucasheriques
    on Oct 16, 2026 – with Claude
    Contributor

    Team MCP Analytics

    Support hero: @gesh (week 1), @lucasheriques (week 2)
    Off during the sprint: @lucasheriques (week 1), @gesh (week 2, Mon–Wed)

    OKRs

    Details
    • 🛠️ Goal 1: Turn agent feedback into fixes
    • 🚀 Goal 2: Prepare MCP analytics for GA (stretch)
    • 🌱 Goal 3: Help more teams start using MCP analytics
    • 🧪 Goal 4: MCP evals

    See all goals in detail here

    Retro

    Details ⚪ = Not Started 🟡 = In Progress 🟢 = Completed 🔴 = Won't complete 🔵 = Abandoned 🟣 = Ongoing

    @lucasheriques

    • 🟢 Support hero
    • 🟢 Wrap up the Go SDK: release it and set up the wizard for it
    • 🟣 MCP evals: land the agent eval runner for any MCP server and run it against our own MCP server
    • 🟢 Docs page for using MCP analytics without SDK support (generic event capture)

    @gesh

    • 🟢 Wrap up the WebMCP extension + add docs
    • 🟢 posthog.com WebMCP capturing works smoothly and events follow the SDK schema
    • ⚪ Agent feedback UI
    • 🟢 Dashboard revamp (stable AI lab colors, charts grouped by lab, absolute values)
    • 🟢 Logos revamp
    • 🟡 MCP analytics events plan

    @mjwarren3

    • 🟢 Pricing RFC

    Plan

    @lucasheriques

    • Activation: add MCP analytics to the general wizard onboarding
    • Activation: in-app onboarding for MCP analytics (part of the GA plan)
    • Make sure the docs are up to date
    • MCP evals (ongoing)

    @gesh

    • Wrap up MCP analytics events plan
    • Make MCP analytics events ready for billing
    • Billing (might not be able to complete all):
      • New billing plans
      • Usage report
      • Quota limiting
      • Pricing calculator (website)

    @mjwarren3

    Generated by Claude Code

  12. skoob13 commented on Oct 16, 2026

    @skoob13
    skoob13
    on Oct 16, 2026
    Contributor

    Team Self-Driving

    Support hero: @andrewm4894 @mikaylathompson
    Off during the sprint:

    Quarter goals

    Team objectives. Quarter-level status to confirm; this update covers the sprint commitments and their progress.

    Retro

    Tasks

    High priority

    • 🟢 Offsite: Team planning and collaboration.
    • 🟡 Discounts & Trial: Continue inference-cost work and scope/implement the Self-driving free tier. @skoob13
    • 🟢 Tracking costs on Tasks. @oliverb123
    • 🔴 Offline evals and source reliability: Finish reusable environments for long-running agent evals; continue pipeline/source fixes. @sortafreel
    • 🔴 Personal inbox: Dogfood and land the shared selection/order across web, Desktop, and MCP; define success measures before rollout (#110438). @dmarticus
    • 🟢 Report volume and activation: Test dismissal, bulk cleanup, and initial defaults (#110439); run the default scout-creation experiment (#110440); continue interviews and finish outreach setup. @dmarticus
    • 🟢 Ranking and autoresearch: Evaluate model-based inbox ordering; dogfood autoresearch and improve background/rolling scoring (#110079, #110125). @andrewm4894
    • 🟢 Scouts: Start a limited general-scout rollout with cost and quality monitoring; continue model comparisons. @andrewm4894
    • 🟡 Close the loop: 🟢 Finish report monitoring and inconclusive outcomes, then dogfood the complete measurement → implementation → 🟢 monitoring → verified-result flow. @mikaylathompson @andrewm4894
    • 🔴 Continue Jev evaluations before deciding on rollout. @mikaylathompson
    • 🟢 Desktop/mobile: Finish the remaining feedback and usability PRs. @dmarticus

    Low priority / side quests

    • 🔴 First round of scout improvements and additional interviews, if capacity permits. @sortafreel
    • 🟡 Explore per-scout MCP tool grants (#108530) and write about skills. @andrewm4894

    Plan

    High priority

    • [Mobile] Launch PostHog mobile. Start with internal dogfooding. Get it into both app stores and either get it launched into the public app stores or, at minimum, get a publicly accessible beta version. @cvolzer3

    • [Mobile] Get with Alex V and launch V1 of support in PostHog mobile @cvolzer3

    • [Mobile] Do customer interviews on the mobile use cases in self-driving and agent-mediated workflows @cvolzer3

    • [Mobile] Work on alerts (configurable notifications for the mobile app) and tie that back into the alerts platform @cvolzer3

    • [Platform] Iterate on the MCP tooling necessary for external agents to interact with reports @cvolzer3

    • [General] Iterate on the core reports experience from a mobile & agent-mediated perspective @cvolzer3

    • Inbox ranking: kick off the default-sort experiment (control vs open, 4 weeks), check the first prod nn training run, then project vectors and after that user vectors for more personalization @andrewm4894

      • get project vectors and user vectors flowing and get NN serving working
      • scale and improve the model a bit - ml stuff

    • Report checks make sure operationally working well and help see how can fit or be part of monitoring work if useful. @andrewm4894

    • Scouts pre-checks: confirm the error tracking, replay and replay vision pre-checks actually run #115934, #115933, #115935, then logs, apm, insight alerts and the data ones @andrewm4894

    • Scouts: retire the revenue analytics scout by Oct 22 #117599, auto-pause silent configs, tool access enforcement and settings UI @andrewm4894

    • Datadog inbox source: land #113615, sync with the author on Oct 23 @andrewm4894

    • Inbox scopes step 1 #117496, unblocks user-authored reports @andrewm4894

    • Get dogfooding scout trials etc @andrewm4894

    • Keep doogfooding on personal scout stuff - once pricing lands want to be able to scaffold or help users set up personal scouts if they want. @andrewm4894

      • might be a few small features needed for this

    • Ship billing side integration (and some fixes), so we can turn on new billing with a feature flag (70% of the way there, got an agent noodling on some details) @oliverb123

    • Triage our security tickets - this slipped, needs to be done, will probably need to pull folks in depending on areas @oliverb123

    • Start working on pipeline the thirdening @oliverb123

    • Set up the onboarding folks so they're good to go with the customer account audit stuff (need to provision some credentials) @oliverb123

    • Need to merge monolith PR to use that CH user - open, stamped and green, but a bit of a risky PR, so will wait till monday morning @oliverb123

    • Work with agent-infra & billing teams on overall AI pricing topics (doc in progress, deal with feedback next week) @annikaschmid

    • check customers on new pricing & set up some calls @annikaschmid

    • growth review @annikaschmid

    • Close the loop @mikaylathompson

      • backend: metric checks running scaled checks throughout monitoring period
      • backend: push towards metric checks over agentic checks; push towards shorter windows, better thresholds
      • backend: automatically moving to resolved/opening new report based on check status
      • frontend: monitoring state view for reports and inbox (while monitoring is in progress, show "looks on track", etc.)
      • mostly frontend: pre-monitoring (ready, implementation) should be fully hooked up to get user buy-in on metrics
      • frontend: exited-out-of-monitoring views for reports
      • frontend: play with a monitoring-only view for the inbox

    • report assignment + report surfaces @dmarticus

    • Interview customers about the Inbox experience + start scheduling calls for user interviews @dmarticus

    • Ship the updated free tier. @skoob13

    • Check other places where we need to change billing info (website, Joe's list, web app). @skoob13

    • Check from the legal perspective whether we need to notify people about changing the free tier. @skoob13

    • Ship the UI-first onboarding. @skoob13

    • Ship the experiment with the new onboarding (GitHub-first version and UI-first version). @skoob13

    • Send an email notification when the org gets a first report. @skoob13

    • Determine the right time for Self-Driving adoption (don't enforce it in regular onboarding). @skoob13

    • Figure out the platform APIs and schedule more interviews with product teams (there is a thread in #dev). @skoob13

    pretty short sprint from me, as I was ill for half of it

    • add improvements for repo selection in Slack bot (use memory) @sortafreel
    • do the first round of "what scouts could be mighty useful for your product" to at least 5 product teams, discuss, merge fixes, create scheduled scouts @sortafreel
    • clean up the scout orchestration/self-improvement toolkit (weird bugs here and there) @sortafreel
    • look into improving scout suggestions (too narrow) @sortafreel

    Low priority / side quests

    Are any other teams impacted by this plan? If so, tag them here

    🟢 =finished 🟡=in progress 🔴=won't finish ⚪=not started

  13. hpouillot commented on Oct 19, 2026

    @hpouillot
    hpouillot
    on Oct 19, 2026
    Contributor

    Team Error Tracking

    Retro 🟢 🔴 🟡 🚫

    @ablaszkiewicz - cymbal path resolution service 🚫 Incident took precedence
    @ablaszkiewicz - error tracking kill switch 🟢 in draft mode
    @ablaszkiewicz - wizard changes for ET setup (faster, more reliable) 🟢

    @fivestarspicy - user interviews 🟢
    @fivestarspicy - pricing / subs analysis 🟡 in progress

    @hpouillot - alerting follow-ups 🟡
    @hpouillot @JamesPatrickGill - Jev based grouping 🟢

    Objectives for the quarter
    • Increase signal, reduce noise
    • Automate issue triage
    • Make Error Tracking first-class for agents
    • Keep Error Tracking bills predictable
    • Fit enterprise workflows

    High priority

    @hpouillot - finish alerting changelog
    @hpouillot - fingerprint table index
    @hpouillot - rollout posthog-js fetch changes
    @hpouillot - rollout severity inference and grouping

    @ablaszkiewicz - cymbal path resolution service

    @JamesPatrickGill - Issue summary
    @JamesPatrickGill - side quest: Issue tracker improvement / fix
    @JamesPatrickGill - side quest: Cymbal latency
    @JamesPatrickGill - side quest: APM discussion

    @fivestarspicy - pricing RFC improvements

  14. thmsobrmlr commented on Oct 19, 2026

    @thmsobrmlr
    thmsobrmlr
    on Oct 19, 2026 – with Claude Github MCP Connector
    Collaborator

    Team Product Analytics

    Support hero: @MattPua
    Off during the sprint:

    Retro

    Details

    High priority

    • 🟢 September growth reviews @mjwarren3
    • ✅ Complete 10 user interviews with frequent MCP users, present/analyze findings and figure out what users need @sampennington
    • ✅ Monitoring and alerts for our MCP usage @sampennington
    • 🔄 Make charts/dashboards created from MCP match the question more reliably @sampennington
      • Improved dashboard skill pick up rates by quite a bit
      • Insight creates using properly query runners more reliably after follow up Q's PR
    • 5-10 User interviews with users who have churned (invites to active org & completely fresh users) @GeneralistDev
    • Product Analytics dashboard experiment @GeneralistDev
    • Capture 10 key product analytics agent question workflows into reliable evals and do targeted skill updates @GeneralistDev
    • ✅ Figure out how to contribute to all of this @thmsobrmlr

    Low priority / side quests

    • 🔄 Kill unused features @sampennington
      • Correlation analysis deletion PR is up, just pending customer-analytics agreement
    • Streamline the pipeline for writing, testing (evals) and releasing skills, all the bits are there, just not working well @GeneralistDev
    • Finish started work from Q3 @thmsobrmlr
    • 🟡 Redesign the Views tab in the new UI @mjwarren3

    Plan

    High priority

    • playing catchup from PTO @MattPua
    • analytics/MCP user interviews sent before PTO @MattPua
    • churned small startup investigation/interviews @MattPua
    • Finish data-quality checks @sampennington
    • Ship quill-charts in Canvas's @sampennington
    • Explain this chart @sampennington
    • 5-10 User interviews with users who have churned (invites to active org & completely fresh users) @GeneralistDev
      • Emails have been sent, not responses yet
    • Capture 10 key product analytics agent question workflows into reliable evals and do targeted skill updates @GeneralistDev
      • Eval harness improvements had to come first
    • Reach out to third party artifact users for user interviews @thmsobrmlr
    • Dig through BuildBetter and Gong for market trends; Create survey for customer facing folks @thmsobrmlr
    • Redesign the Views tab in the new UI @mjwarren3
    • Get started on big customer discovery, reach out to sales team @mjwarren3

    Low priority / side quests

    • improve editing UX for dashboards (ive said multiple times, never started, i should start) @MattPua
    • do the all analytics dashboards link from the dashboards list page that Mike prototyped @MattPua
    • Kill more unused features @sampennington
    • Continue with legacy removal, product analytics isolation and show 30-day query usage for properties @thmsobrmlr

    Are any other teams impacted by this plan? If so, tag them here

  15. richardsolomou commented on Oct 19, 2026

    @richardsolomou
    richardsolomou
    on Oct 19, 2026
    Member

    Team Context & MCP

    Support hero: @richardsolomou
    Off during the sprint: @JakeRuth all sprint; @richardsolomou October 28; @adboio October 30.

    Quarter goals

    Goals

    • 🟡 Goal 1: Prove or disprove the context layer's value.
      • 🟡 Interview 20 external users and five internal users, then synthesize the findings.
      • ⚪ Write a high-level RFC for the context layer.
      • ⚪ Document how the context layer could unify Business Knowledge, the semantic layer, and other PostHog context systems.
    • 🟡 Goal 2: Choose the right MCP architecture.
      • ⚪ Prototype and compare promising approaches, including code execution and the current tool-based architecture.
      • ⚪ Recommend an architecture based on the findings.
      • ⚪ If the current architecture remains the best option, update the objective to focus on improving it.
    • ⚪ Goal 3: Define metrics to measure the team's impact.
      • ⚪ Define a small set of revenue and/or usage metrics.
      • ⚪ Document the calculations and establish baselines.
      • ⚪ Set up a dashboard and a regular review cadence.

    Retro

    Details

    @richardsolomou

    • 🟢 Published the team's Q4 investigation goals.
    • 🟢 Automated approval and merge of AI-plugin skill sync PRs.
    • 🟢 Changed Cloudflare MCP builds to install only MCP dependencies.
    • 🟡 Prepared the ChatGPT connector refresh for missing permissions; OpenAI review was pending at sprint end.
    • 🟡 Tightened MCP connection policies and response limits; merged October 17, after the sprint.
    • Side quests:
      • 🟡 Fixed unnecessary source-map prompts for mapped Convex errors; the upstream PR remains open.

    @adboio

    • 🟢 Built native interview invitations, including Codex support and recruitment diagnostics.
    • 🟢 Added full-text catalog search.
    • 🟢 Enabled saving event-table insights through MCP.
    • 🟢 Added a flagged trial asking agents to describe their goal before calling tools.
    • 🟡 Built the context-selection prototype for web and Slack; still in review.
    • 🟡 Built filtered event subscriptions and previews for MCP; still in review.
    • 🟡 Investigated growing MCP query latency; no confirmed fix yet.
    • Side quests:
      • 🟡 Started removing retired Product tours from PostHog and the SDK.
      • 🟢 Fixed credentials used for PR screenshot and video uploads.
      • 🟢 Improved guidance for explaining the problem in PR descriptions.

    @JakeRuth

    • 🟢 Fixed default project selection for organization-scoped MCP tokens.
    • 🟢 Mapped MCP clients and their differing behavior to inform architecture work.
    • 🟢 Investigated Cloudflare build failures and documented the findings.
    • Side quests:
      • 🟡 Built the Club Hoguin prototype; hosting and release remain unfinished.

    🟢 =finished 🟡=in progress 🔴=won't finish ⚪=not started

    Plan

    High priority

    @richardsolomou

    @adboio

    @JakeRuth — off throughout.

    Low priority / side quests

    Are any other teams impacted by this plan? If so, tag them here

    • @PostHog/team-workflows — event subscriptions and delivery.
    • @PostHog/team-agent-infrastructure — context-selection runtime and Tasks integration.
    • Product teams — ownership of tool-health findings and skill publishing guidance.

  16. Radu-Raicea commented on Oct 19, 2026

    @Radu-Raicea
    Radu-Raicea
    on Oct 19, 2026
    Member

    Team AI Observability

    Retro

    Details

    Last sprint was half in the mixed team offsite, then 1 day off for the engineers.

    Plan (W43, W44)

  17. TueHaulund commented on Oct 19, 2026

    @TueHaulund
    TueHaulund
    on Oct 19, 2026
    Contributor

    Team Session Replay

    Support hero:

    @ksvat

    OKRs

    🔬 Goal 1: Refine the core of Replay Vision
    🌍 Goal 2: Replay Vision everywhere, but better
    🔗 Goal 3: Vastly improve the integration between Session Replay and Replay Vision
    🤝 Goal 4: Stay close to our customers
    ❤️ Goal 5: Show Session Replay some love

    Retro ✅ 🔴 🟡 🚫

    ✅ @ksvat Finalize, release and test experiment-scanners

    ✅ @fasyy612 Build and test what-to-watch prototype for Session Replay
    ✅ @fasyy612 Kill unused Session Replay features

    🟡 @fivestarspicy Cost-tagging crypto shredding
    🟡 @fivestarspicy Follow up w/ event team wrt customer-meeting @ small team offsite in SF
    ✅ @fivestarspicy Set up scout to look for interesting or novel scanners

    ✅ @TueHaulund Implement programmatically triggered observations

    ✅ @ksvat Follow up on Jev ranking the what-to-watch tab experiment
    🟡 @TueHaulund Follow up on labelled-data test suite once v3 recordings are labelled
    ✅ @TueHaulund Update scanner creation flow and rerun experiment

    High priority

    @ksvat Get started on funnel scanners
    @ksvat Figure out how to kill old experiment targetting

    @fasyy612 Generate examples for Session Replay UI rework to iterate on (dev mode vs simple mode)
    @fasyy612 Continuously follow up on W2W tab results and iterate on empty state

    @fivestarspicy Update cost-tagging
    @fivestarspicy MCP usage dashboard
    @fivestarspicy Follow up w/ event team wrt customer-meeting @ small team offsite in SF

    @TueHaulund Meet with Google sales team and get us started on the path to a GCMA and ZDR

    Side quest

    @TueHaulund Prep for RV promo event in CPH
    @ksvat Prep for RV promo event in LA
    @TueHaulund Prep for small team offsite in SF

    @fivestarspicy PMF survey for Session Replay and Replay Vision
    @fivestarspicy Brainstorm potential new scanner types after funnel scanner

  18. lricoy commented on Oct 19, 2026

    @lricoy
    lricoy
    on Oct 19, 2026
    Member

    Team Web Analytics

    Legend
    ⚪ = Not Started 🟡 = In Progress 🟢 = Completed 🔴 = Won't complete

    Support hero

    @jordanm-posthog

    OOO 26-6th Nov
    OOO Arthur travelling back

    Retro:

    Details

    High priority

    🟢 @arthurdedeus unalive vitally
    🟡 @arthurdedeus person group membership
    🟢 @arthurdedeus account details pages' new widgets
    🟢 @arthurdedeus tackle issues from the onboarding team

    🟢 @jordanm-posthog historical heatmaps variants
    🟡 @jordanm-posthog launch AEO (alpha tag)
    🟢 @jordanm-posthog cost observability for services

    🟢 @jabahamondes integrate the search table with posthog conversion goals
    🟢 @jabahamondes onboarding update -> suggest paid ads sources that are already being user in events
    🟡 @jabahamondes suggest better conversions goals based on events

    🟢 @mjwarren3 marketing analytics pricing RFC

    🟡 @lricoy launch AEO (alpha tag)
    🟡 @lricoy support work for breached escalated tickets
    🟢 @lricoy move very large EU customer

    High priority

    @jordanm-posthog and @lricoy launch AEO (alpha tag)
    @lricoy backfill sessions v3
    @lricoy run our eval list of prompts against wizard pages

    @arthurdedeus work on reliability of CA
    @arthurdedeus nail email matching logics after last sync
    @arthurdedeus person-group membership

    @jabahamondes get recommendations for google Ads + Search console data
    @jabahamondes business recommendations mcp for google ads

    @mjwarren3 finalize Marketing Analytics pricing RFC
    @mjwarren3 do-alive customer analytics end user product

  19. charlesvien commented on Oct 20, 2026

    @charlesvien
    charlesvien
    on Oct 20, 2026
    Member

    Team Agent Infrastructure

    Support hero: Alessandro first week. Charles covering Brandon’s rotation while he’s away in second week.
    Off during the sprint:

    • Brandon is OOO from Oct. 26 to Nov. 7
    • Charles is OOO on Oct. 23
    • Alessandro is OOO Oct. 26 + Oct. 28 (conf)

    Quarterly goals

    1. 🟢 Reduce the complexity of building AI products (Driver: @charlesvien)
    2. 🟢 Every AI unit of work is costed and attributed (Driver: @brandonleung)
    3. 🟢 Safe model and sandbox access (Driver: @tatoalo)
    4. 🟢 A reliable agent runtime teams can build on (Driver: @jonathanlab)

    https://posthog.com/teams/agent-infrastructure

    Plan

    High priority

    • @charlesvien: ship TaskProfile implementation, keeping this focused on the initial contract rather than migrating every product
    • @charlesvien: move task status onto Task and migrate existing products
    • @charlesvien: get agent-server run callbacks reviewed and shipped
    • @charlesvien: assist with routing migration to MAP program (cutover schedule ~Nov 9th)
    • @jonathanlab: Simplify code ownership in agent package, simplify/harden agent server
    • @jonathanlab: Figure out biggest wins for agent boot times, get smaller improvements w.r.t. boot time
    • @brandonleung: Migrate replay vision and desktop over
    • @brandonleung: Billing concept for gateway and tasks; carve out org-based limits, caps, ceilings, and wallets while supporting team-based (subset of org)
    • @brandonleung: Implement configurable routing
    • @tatoalo: hogference scaling due to maintain replay vision traffic
    • @tatoalo: agent egress proxy work (this depends on RFC approval tho, external, decision deadline is Oct. 23rd)
    • Support hero (+ team): triage incoming issues and drain the remaining review queue

    Low priority / side quests

    • @jonathanlab: ensure everything is off the legacy gateway, rip out the fallbacks to the legacy gateway in favor of retries to the new gateway, some products are using the legacy gateway as a fallback (phai, workflows, etc)
    • @jonathanlab: hogli tasks:doctor for local dx
    • @jonathanlab: try out smaller agent runtime where we just run in a worker, relying on pi's codemode (if task is only meant for data analysis)
    • @charlesvien: Extract shared-ai alerts that we own into alerts-agent-infra, talk to other teams about them doing the same for their stuff. This large AI group is a bad pattern.
    • @charlesvien: the agent streaming proxy is a orphaned part of our sandbox architecture. It was a quick solution that never got maintained, many clients are still using Django SSE streams.
    • @charlesvien: bring command center back to the new web UI for fun/experiment?
    • @tatoalo: side quests, change sandbox provider, experiment, fun (?)
    • @brandonleung: Clean up our openweight model handling
    • @brandonleung: Documenting agents.md and runbooks

    Are any other teams impacted by this plan? If so, tag them here

    • @PostHog/team-self-driving @PostHog/team-billing: pricing/credits agreement
    • @PostHog/team-replay
    • @PostHog/team-surfaces

  20. haacked commented on Oct 20, 2026

    @haacked
    haacked
    on Oct 20, 2026
    Contributor

    Team Feature Flags

    Support hero:

    Off during the sprint: @haacked: 10/23 and 10/26

    Quarter goals

    Link to goals

    1. Launch realtime cohorts and make behavioral targeting the reason to choose PostHog flags 🟡
    2. Plan for scale and reduce the cost of operating Feature Flags 🟡
    3. Realtime flag definitions ⚪
    4. "Auto" rollback and rollout 🟡
    5. Make flags the default way developers ship end-to-end ⚪
    6. Make flag evaluation correct and consistent with cohorts and Product Analytics 🟡
    Details ⚪ = Not Started 🟡 = In Progress 🟢 = Completed 🔴 = Won't complete 🔵 = Abandoned

    Retro

    Phil continued moving flag calls to flag_evaluations. Every ingestion lane now forks flag calls into the table, every flag view reads from it in mode 1, and insights can chart flag calls from it. The table also stores person columns and properties, collapses duplicate calls, resolves test account filters and the {filters} placeholder, and deletes person properties on request. Flag calls reach destinations in flag_evaluations_mode 2. New action steps and cohort criteria can no longer use $feature_flag_called, and SQL that reads it from events shows a warning. Flags whose dependency failed now fail, and persons DB work in flag evaluation runs under a request deadline. Patricio backfilled legacy secret tokens into project secret API keys, added a kill switch and a shadow check, and turned off legacy team secret token auth. Matheus made cohort processor restarts lossless and gave each pod its own checkpoints, the first steps toward running the processor as multiple pods. Matheus also added 10 teams to the realtime cohort pipeline. Gustavo fixed four flags UI issues, including a warning about a taken flag key while typing, and four cohort bugs. The rollback suggestions RFC is in review. James merged the stale flag health check gate, a variant naming fix, and a rollout timeline fix before leaving the team.

    Details

    @haacked

    flag_evaluations migration:

    Stop advertising $feature_flag_called:

    Personhog client and flags DB timeouts:

    Flag dependency failures:

    HogQL filter fixes:

    Variant key validation:

    Cohorts:

    Replay and remote config:

    Other:

    @gustavohstrassburger

    Flags UI fixes:

    Cohorts fixes:

    In progress:

    @JamesPatrickGill

    Flags fixes:

    Infrastructure:

    @matheus-vb

    Multi-pod cohort processor:

    Realtime cohort rollout:

    Other:

    @patricio-posthog

    Secret API key migration:

    Other:

    Plan

    Project Board

    High priority

    @haacked

    @gustavohstrassburger

    @matheus-vb

    @patricio-posthog

    Side quests

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    sprintSprint planning

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions