# Source Pack: Portal Event 68 - Jake Winckowski Fireside
## Step Status
Source-pack complete. This pack uses verified Portal event data plus Prism transcript and summary artifacts. It separates direct session evidence from outside knowledge and records open gaps for later angle planning.
## Source Ledger
### Portal Event
- Source: Payload CMS / Portal event record
- Portal event id: 68
- Public URL: https://portal.raidguild.org/events/68
- Title: June Cohort Fireside Chats (Jake Winckowski)
- Summary: June Cohort Fireside Chat with Jake Winckowski. Calendar: RaidGuild. Created by: clerics@raidguild.org
- Session type: brownbag
- Starts: 2026-06-23T17:00:00.000Z
- Ends: 2026-06-23T17:30:00.000Z
- Location: RaidGuild Discord
- Discord event: https://discord.com/events/684227450204323876/1516478200660758660
- Discord scheduled event id: 1516478200660758660
- Source status: summarized
- Related Portal thread: How to RaidGuild - Field Experience from the Edge (thread id 5, slug how-to-raidguild-field-experience-from-the-edge)
### Prism Transcript Artifact
- Artifact id: 20260623_173330Z-discord-voice-c37cd3ce
- Human URL: https://prism-memory-production-002c.up.railway.app/artifacts/20260623_173330Z-discord-voice-c37cd3ce
- Source: discord-voice
- Type/status: meeting_transcript / processed
- Created: 2026-06-23T17:33:30Z
- Recording/session id: 9058b934-6970-4aaa-8895-ba4b3a7d3466
- Voice channel: Discord #cohort-voice
- Started: 2026-06-23T17:00:10.802Z
- Ended: 2026-06-23T17:33:30.941Z
- Participants listed by artifact: duckanbro, 0xHunter, Aphilos • Pharo, chickenparm, ECWireless, louchi, samkuhlmann
- Content length: 34038 characters
### Prism Summary Artifact
- Artifact id: 20260623_173330Z-discord-voice-4e530776
- Human URL: https://prism-memory-production-002c.up.railway.app/artifacts/20260623_173330Z-discord-voice-4e530776
- Source: discord-voice
- Type/status: meeting_summary / processed
- Created: 2026-06-23T17:33:30Z
- Title: AI-Augmented Customer Experience at HackerOne
- Tags listed by artifact: ai-customer-support, hackerone, claude-code, human-in-the-loop, support-automation, slack-app, ai-adoption
- Content length: 5331 characters
## Direct Session Evidence
### Session Frame
The host framed the conversation as part of a Fireside Chats series gathering perspectives on how people in different industries and roles are adapting to AI. The host also stated that audio was being recorded for transcript ingestion and later content/research use.
Jake asked whether the conversation would be posted publicly and said he was trying to decide how much he could share, while avoiding vulnerability data. Treat this as a public-output caution: later drafts should avoid confidential vulnerability, customer, or internal security details unless they are explicitly present in already-approved public material.
### Speaker Context
Jake introduced himself as someone who runs customer experience at HackerOne and described himself as business/customer-experience oriented rather than a developer by trade. He said he had some technical background from working at Apple but that the development world was relatively new to him.
Direct evidence supports identifying the guest as Jake / chickenparm in the transcript. The Portal title says Jake Winckowski. The artifacts do not independently provide a surname beyond the Portal title.
### Core Topic
The session centers on AI-augmented customer experience operations at HackerOne: increased ticket volume, internal support tooling, Slack-based workflows, Claude Code, data access, confidence scoring, security constraints, and human-in-the-loop support automation.
### HackerOne / Bug Bounty Context
Jake described HackerOne as a bug bounty / vulnerability disclosure platform where companies can set up vulnerability disclosure programs or paid bug bounty programs. He gave examples of program scope and hacker payouts. This should be handled carefully in public copy: do not turn conversational examples into current verified market facts without outside validation.
### AI-Driven Volume Increase
Jake said hackers are building agents to find and submit more issues, and that support ticket volume increased sharply because AI makes it easier for users to submit requests. The summary artifact captures this as ticket volume going from an unspecified baseline to roughly three times as much overnight in one area.
Evidence confidence: high for the claim that Jake reported a sharp increase; low for any exact quantified metric because the source says "X" and "three X" conversationally, not a precise audited number.
### Dispatch / Slack App Workflow
Jake described building an internal Slack app referred to in the transcript as "Disfaction" and in the summary as "Dispatch." The system sends tickets into a team Slack channel and supports buttons/actions such as approve, approve-and-resolve, edit, and escalate. It suggests responses by drawing from Freshdesk tickets, GitLab issues, Linear issues, documentation, and resolved cases.
Gap: the app name is inconsistent between transcript and summary. Use "the internal Slack support app" unless a human confirms the exact name.
### Human In The Loop
Jake said the team is not letting automation respond directly yet because he still has trust concerns. Current use is human review of suggested responses. The intended path is to automate high-volume, low-risk ticket categories once confidence becomes high enough.
Evidence confidence: high. This is repeated in the transcript and summary.
### Confidence Scoring
Jake described confidence scores as an active tuning area. He said scores improved as more historical data became available, including a large GitLab issue history, but still need work when a response is mostly correct yet contains one section that cannot be sent.
Evidence confidence: high for the qualitative pattern. Avoid exact numeric claims unless framed as Jake's rough observation during the session.
### Operational Improvements
Directly supported improvements include:
- less context switching from Freshdesk to Slack-centered work
- ticket properties filled automatically
- ability to tag AI-assisted tickets for reporting
- faster internal tooling changes, including adding an approve-and-resolve button in minutes rather than waiting on a formal development request
- support agents spending more time on deeper issues after lower-effort issues are handled faster
### Security And Data Access
Jake described a shift from heavy AI restrictions to broader approved usage after the company became more comfortable with security controls. He also described data classification policies and approved environments. The biggest current blocker he named was data access: no API key, not being allowed to use a key, or no API existing.
Evidence confidence: high for Jake's statements about his experience. Do not infer HackerOne-wide policy details beyond what Jake said.
### Rate Limits And Background Fetching
Jake described hitting Freshdesk API rate limits, learning about caching, and moving some historical fetches or refreshes to scheduled background jobs during off-hours.
Evidence confidence: high.
### Support Role Outlook
Jake did not present a purely optimistic or pessimistic view. He said human-in-the-loop work will remain important for the next few years, especially for areas like payment issues, mediation, technical troubleshooting, and judgment-heavy cases. He also described a three-month upskilling program for support agents to build more technical capability.
Evidence confidence: high.
## Outside Knowledge
No outside factual research was added in this step. The source pack is grounded only in:
- Portal event 68
- Prism transcript artifact 20260623_173330Z-discord-voice-c37cd3ce
- Prism summary artifact 20260623_173330Z-discord-voice-4e530776
- prior intake artifact on request #240
Later steps may optionally validate public HackerOne facts, Claude Code naming, or product references before public-facing copy.
## Gaps And Uncertainties
- Exact app name: transcript appears to say "Disfaction" while the summary says "Dispatch." Confirm before naming it in public copy.
- Public approval boundary: Jake asked about whether the session would be posted publicly and indicated he would avoid vulnerability data. Later public outputs should be reviewed for private/security-sensitive details.
- Speaker identity mapping: Portal title gives Jake Winckowski; voice artifacts identify participant handle chickenparm. The mapping is likely but should be treated as source-derived from Portal + transcript context, not independently verified elsewhere.
- Company/product claims: HackerOne platform details, Coinbase payout example, Department of Defense reference, Claude Code/Bedrock setup, and internal security/data classification practices are all session statements, not externally validated facts.
- Metrics: "three X overnight," confidence-score movement, and development-time comparisons are conversational estimates. Use as anecdotes, not audited metrics.
- Output scope: intake noted no human narrowed scope to session-report-only versus wiki/blog/child-request candidates. Continue with the registered full workflow unless a later gate narrows it.
## Candidate Follow-Up Questions
1. Can Jake confirm whether the internal app should be called Dispatch, Disfaction, or something else?
2. Which parts of the session are approved for public quoting or paraphrase, given the initial public-posting question?
3. Should HackerOne company facts be externally verified before a blog/wiki candidate is drafted?
4. Are the most useful public angles operational, cultural, security/process, or workforce/upskilling focused?
5. Should later outputs avoid naming internal systems such as Freshdesk, GitLab, Linear, Bedrock, or Slack unless they remain high-level and source-framed?
## Candidate Angle Seeds For Next Step
These are not final angles, only source-grounded seeds for angle-plan:
- Non-developers building internal tools with coding agents: Jake used Claude Code to build support tooling despite not being a developer by trade.
- Human-in-the-loop as an adoption bridge: automation is useful now, but reviewed suggestions and confidence scoring make it operationally acceptable.
- Data access is the real ceiling: the agent can only help where approved APIs, permissions, and source systems are available.
- AI changes both sides of support: agents increase inbound volume while also helping teams process that volume.
- Upskilling instead of replacement panic: support agents can move toward technical troubleshooting and higher-judgment work.
## Recommended Next Step
Advance to angle-plan. Use this source pack to build topic candidates and content angles, with explicit public-safety flags around confidential security/vulnerability details and unverified company claims.