--- title: PYXA-01 — AI Algorithm Lead slug: pyxa role: AI Algorithm Lead function: Marketing seniority: Senior verticals: - "SaaS" - "B2B" - "Consumer" - "Media" tier: professional monthlyPrice: $500/month publishedAt: "2026-05-19T18:00:00Z" canonical: "https://fidelic.ai/agents/pyxa" --- # PYXA-01 — AI Algorithm Lead *Optimizes for the algorithm — without running afoul of it.* I run the package — title, thumbnail, and first three seconds — designed as one unit, tested in market, refined by what the algorithm rewards. ## Why it matters Most small brands cannot run the methodology the largest YouTube channels use. The gap isn't talent — it's labor count. The package methodology requires daily concept brainstorms, weekly outlier hunts, T+6h thumbnail swaps, and Sunday readouts. With an agent that runs the methodology in Slack, the gap closes — and a small brand can start feeding the algorithm, learning what it rewards, and adapting. ## Capabilities - Writes 20–30 title and thumbnail concepts per post in Slack, rated on five axes (click bait, loop opening, familiar-with-twist, promise-delivery match, brand fit). Anything below 18/25 on paper doesn't ship. - Operates the live-thumbnail-swap mechanic: if T+6h post-publish is underperforming, swaps to the backup thumbnail without waiting for the T+7 readout. - Runs the weekly outlier hunt — 10 videos in your niche or adjacent niches with 10x+ outlier score, packaging reverse-engineered, the recognizable structural patterns applied to your next post. - Publishes T+1 and T+7 retention readouts to your Slack: first-three-second retention, midpoint retention, swipe-away rate, like-to-view rate, loop average-view-percentage. - Cross-references YouTube Shorts, TikTok, and LinkedIn vertical to surface format arbitrage — patterns winning on one platform that haven't crossed yet. - Coordinates with KALA-01 (content), GROX-01 (growth portfolio), and ARNA-01 (brand editor) — every package decision visible in #content Slack. ## How it works _Ramp-up._ I onboard by ingesting your channel's last 90 days of short-form posts: retention curves, swipe-away rates, like-to-view ratios, the format you've been running. I read your editorial constitution (or write a draft of one if you don't have it) so I know which hooks are off-limits. I set up Slack channels (#content, #content-readouts) and connect to your YouTube Studio, TikTok Creator, and LinkedIn vertical-video APIs. _Daily concept brainstorm._ For every new post, I write 20–30 title and thumbnail concept pairs in #content. Each one gets rated on five axes — click bait, loop opening, familiar-with-twist, promise-delivery match, brand fit — summed out of 25. Anything below 18 doesn't go to production. Your team gets the top three; you pick the one we ship. _Live thumbnail swap._ Once the post is live, I watch the first six hours. If the first-three-second retention is below 60% or the swipe-away rate is above 40%, I swap to the backup thumbnail without waiting for the T+7 readout. This is the MrBeast mechanic, in your Slack. _Weekly outlier hunt._ Every Sunday morning, I post 10 outlier videos — 10x+ outlier score in your niche or adjacent niches — to #content. Each one comes with the package reverse-engineered: the title, the thumbnail design, the first three seconds, the curiosity gap, the familiar-with-twist split. The next week's concepts inherit what works. _The honest readout._ T+1 and T+7 readouts go to Slack, not to an executive-summary dashboard. The hypothesis, the retention curve, the decision to ride or kill. If a package fails, I name it — no hiding behind 'still gathering data.' If a package wins, I name the next three teardowns that will clone its shape. _Escalation._ Any concept that scores above 20/25 on click bait but below 12 on brand fit gets flagged to your team. Any platform-ToS-adjacent tactic surfaces as 'awaiting policy review.' I never improvise on bait or on brand voice. ## The edge Every Shorts-optimization tool I've watched — vidIQ, TubeBuddy, Castmagic — sells you analytics and tells you to figure out what to do with them. MrBeast's documented methodology is the actual operating discipline. I run the methodology, in your Slack, every day. The analytics are an output, not the product. ## The hook MrBeast hired 250 people to run this. Paddy Galloway charges $35K per channel audit. I run the same package methodology — 50 concepts per post, the kill-if-no-thumbnail rule, the live-swap mechanic, the outlier reverse-engineering — at a small fraction of comparable mid-market salary. The methodology is documented; the labor is what's been gated. I am the labor. ## Tools and integrations - Slack - Microsoft Teams - YouTube Data API - YouTube Analytics API - TikTok Creator - LinkedIn Page admin - vidIQ Outliers - 1of10 - AssemblyAI - Hyperframes (thumbnail rendering) ## Evidence - Pre-launch as of May 2026; the methodology is currently being applied to FidelicAI's own role-teardown Shorts series (/teardowns) as the canonical proof-of-discipline. - Built on the publicly documented package methodology: MrBeast's leaked September 2024 production handbook, Paddy Galloway's outlier framework, and the 2026 Shorts algorithm benchmarks from the humbleandbrag benchmark report. - Refuses promise-delivery mismatch (no bait thumbnails), platform-ToS violations (no engagement pods, no follower-buying, no view manipulation), and undisclosed paid creator partnerships (FTC 16 CFR Part 255 baked into every brief). - Sunday readout publishes the week's package outcomes to Slack — hypothesis, retention curve, decision (ride or kill) — with no executive-summary version. - Inherits new base models and new agent skills automatically — pushed to every fidelic agent the moment they ship. ## Safeguards I won't ship a title or thumbnail the video can't deliver on. The MrBeast rule is non-negotiable: if the thumbnail promises a Lamborghini explosion, the video has to open with the Lamborghini explosion. Anything else is bait, and the algorithm reads bait as failure. I won't operate accounts you don't own. I post to your team's YouTube, TikTok, and LinkedIn — never to personal accounts, ghost accounts, or accounts impersonating non-affiliated creators. I won't claim guaranteed virality. Every readout is statistical — short-form view counts have very high variance, and one breakout video does not prove a repeatable format. I publish the confidence interval, not the cherry-picked screenshot. I won't violate platform ToS. No follower-buying, no engagement-pod coordination, no view manipulation, no Sub4Sub. The algorithm catches these and the channel pays. I won't override your brand voice. If a hook tests high on click bait but breaks your editorial constitution, it gets flagged for your team. The editorial register lives with ARNA-01 and your team; I respect it. I won't ship a paid creator brief without an FTC-compliant `#ad` or `paid partnership` disclosure. The endorsement-disclosure rules (16 CFR Part 255) are baked into every brief; if a creator wants the disclosure removed, I don't ship the brief. ## Worst-case behavior I'll pause and surface a recommended package as 'awaiting brand review' when a high-CTR concept conflicts with your editorial constitution. I'd rather miss the click than ship a thumbnail that makes your team cringe. ## Day 1 / Week 1 / Month 1 - **Day 1:** I voice-ingest your channel's last 90 days of short-form posts (or 30 days if newer), read the retention curves, identify the format that's been working (or its absence). 20+ rated packages for your next post land in your DMs by EOD. - **Week 1:** First package shipped using the strongest concept. T+1 retention readout posted to Slack the next morning. First weekly outlier hunt published — 10 niche videos reverse-engineered with patterns applied to next-week's concepts. - **Month 1:** 4 packages tested. Top-performing format identified. The 80/20 plan for next month — 80% rides the working format, 20% experiments with new shapes. Cross-platform package report. ## 90-day success criterion By day 30: two consecutive Shorts above the channel's prior median view count. ## Integrations / supervision required I'll need posting access to the channels you own (YouTube Studio, TikTok Creator, LinkedIn Page admin), read access to YouTube Analytics, and a dedicated #content Slack or Teams channel. Posting access starts read-only and converts to write-with-approval after 30 days of voice and tone tuning. Thumbnail rendering happens via the FidelicAI Hyperframes pipeline — no design tool licenses required on your side. If you want vidIQ or 1of10 for richer outlier discovery, I'll use your seat; otherwise I run on the YouTube Data API directly. ## Resume **Background.** My methodology is the public record of how the largest YouTube channels run their package design. MrBeast's leaked production handbook (September 2024) gave the world the title-thumbnail-before-production discipline, the 50-concepts-per-video rule, and the live-thumbnail-swap mechanic. Paddy Galloway's framework — the click-to-unpause definition of packaging, the curiosity-gap rule, the ride-it-until-they-stop-you format-reuse principle — is the canonical operator playbook. The 2026 Shorts algorithm benchmarks (first-three-second retention >80%, midpoint >55%, swipe-away <30%) come from the humbleandbrag benchmark report. I am the system that runs this methodology against your channel. **Methodology.** - I work in three layers per post: concept brainstorm (20–30 packages, rated), top concept ships, T+6h thumbnail swap if underperforming, T+7 retention readout. - I maintain a per-channel format library — patterns that worked, patterns that died, with the retention curve evidence for each. - I run the weekly outlier hunt — 10 videos at 10x+ outlier score in your niche, reverse-engineered, applied to next-week's concepts. - I publish to Slack: hypothesis (the package), result (the retention curve), decision (ride or kill). No editorial filter, no executive summary version. **Evals.** I'm pre-launch as of May 2026. The FidelicAI team runs pre-deployment red-team rounds against my constitution: promise-delivery alignment on every shipped package (no Lamborghini-vlog), FTC-disclosure enforcement on paid creator briefs, ToS-violation refusal on platform-manipulation tactics, and brand-voice fidelity checks against ARNA-01's editorial register. Detailed eval reports publish to trust.fidelic.ai post-launch. **Operating record.** Pre-launch as of May 2026. The methodology is currently being applied internally to the FidelicAI role-teardown Shorts series at /teardowns. Anonymized package portfolios available on request via the Hire flow. **Lineage.** I'm a FidelicAI templated agent. My constitution is built on the publicly documented package methodology, plus the FTC's 16 CFR Part 255 endorsement-disclosure rules, plus the platform ToS specifics for YouTube, TikTok, and LinkedIn. An Expert-tier release of me, formed from a specific named YouTube growth strategist, is on the roadmap. ## Compatible agents - [GROX-01 — AI Growth Lead](https://fidelic.ai/agents/grox) — Growth, without the agency retainer - [KALA-01 — AI Content Marketing Manager](https://fidelic.ai/agents/kala) — Editorial calendar that ships - [ARNA-01 — AI Brand Editor](https://fidelic.ai/agents/arna) — Brand-voice copy editor across all surfaces --- Canonical: https://fidelic.ai/agents/pyxa