◉ TARGET//AGENTIC SYSTEMS · GAMES
MAOR BRIL
PRINCIPAL ENGINEER · AGENTIC AI
You need someone who has already shipped agentic systems for games. I have shipped four.
Plus the multimodal pipeline, the Go SDK and the eval wall underneath them — live at 20M+ monthly users. All 11 of the role's qualifications are scored below, including the ones I reach another way.
04
SHIPPED AGENTIC GAMES
Case Files, StoryLab, c.ai/studio, AI She Wrote — real players, live now
20M+
MAU ON MY AGENTS
AgentX multimodal pipeline: 30+ agentic turns, no human in the loop
11/11
REQUIREMENTS ANSWERED
Scored line-by-line against the role spec — gaps stated, not hidden
58★
OPEN-SOURCE PRIMITIVES
clauder, agentic, JudgeJudy, claude-agent-sdk-go — MIT, in production
01 // THE SCORE
REQUIREMENT MATCH ANALYSIS
// TARGET · NETFLIX JR41530 — MEMBER OF TECHNICAL STAFF, AGENTIC SYSTEMS — GAMES
// THEIR PUBLISHED QUALIFICATIONS, SCORED AGAINST MY RECORD — INCLUDING THE ONES I MISS. TAP ANY LINE.
11/11
ANSWERED
7 DIRECT · 4 ADJACENT
7+ years in AI product strategy, ML and AI engineering, hands-on
○ AWAITING REVIEW
20 yrs shipping · Principal, agentic AI strategy
I own agentic AI strategy at Character.ai and set the org-wide standards for AI tooling — while still writing the systems myself.
Twenty years underneath it: Coinbase, Netflix, Google Cloud, VMware.
▸ TAP FOR EVIDENCE · +4% SYNC
3+ years of experience in the game development industry
○ AWAITING REVIEW
4 shipped products, real players. No studio badge.
My route was not a game studio, and I would rather say so than stretch a date range.
Case Files — timed detective game, Character.ai Interactive Originals. StoryLab — LitRPG audio. c.ai/studio — weekly microdrama. AI She Wrote — built and operated alone.
If the requirement is studio tenure, I do not meet it. If it is shipping games that agents run, I do.
▸ TAP FOR EVIDENCE · +4% SYNC
Product management: user needs, roadmaps, production workstreams
○ AWAITING REVIEW
Owns the roadmap. From the builder's seat, not a PM's.
I define what gets built for agentic AI at Character.ai and run the workstreams that ship it, which is the 'Builder PM' half of this role. What I have not done is carry a formal PM title, run structured user research, or own a roadmap negotiated across a product org. I have prioritized by shipping and reading telemetry, not by writing PRDs.
▸ TAP FOR EVIDENCE · +4% SYNC
Games or interactive entertainment — shipped features, game engines (Unreal, Unity), or AI experiences for players
○ AWAITING REVIEW
AI for players: 4×. Unreal/Unity: no.
This requirement is an or, and I clear the third branch decisively — building AI experiences for players is the whole archive above. I have not worked in Unreal or Unity, and I am not going to imply otherwise. My interactive surfaces are web and app clients over agent backends. If the work is engine-side gameplay programming, that is a ramp for me; if it is the agent layer that talks to the engine, that is what I already build.
▸ TAP FOR EVIDENCE · +4% SYNC
Deep, practical agentic AI in production — multi-step reasoning, tool use, multi-agent orchestration
○ AWAITING REVIEW
AgentX — 30+ agentic turns, 20M+ MAU, no human in the loop
A planner agent decomposes the request, executor agents generate image, video, and audio, and reviewer agents gate every stage on trained eval scores. 30+ agentic turns, 6 specialized tools, no human in the loop after the user submits a prompt.
▸ TAP FOR EVIDENCE · +4% SYNC
Strong Python and production experience with agentic frameworks (LangChain, LangGraph, AutoGen, Google ADK)
○ AWAITING REVIEW
Python for post-training and evals. Go for the runtime, by choice.
Evals, judge training and the annotation loop are Python.
The runtime is not. I open-sourced claude-agent-sdk-go instead of adopting LangChain — at 20M+ MAU I wanted bounded goroutine pools, real context propagation, a static binary.
I know those frameworks and chose against them. A house language is a preference, not a skill gap.
▸ TAP FOR EVIDENCE · +4% SYNC
Designing and operating evaluation infrastructure — offline benchmarks, automated scoring, online experimentation
○ AWAITING REVIEW
JudgeJudy + Judge Dredd — offline harness, online loop
Open-sourced a multimodal eval harness (AI judges, trained scoring models, human scoring pipeline), then built the self-improving layer: Judge Dredd pulls Amplitude and Statsig telemetry over MCP, correlates eval scores against real user behavior, and feeds validated gates back into the runtime. Every AgentX release clears the wall before players see it. This is the requirement I would point at first.
▸ TAP FOR EVIDENCE · +4% SYNC
Deep GenAI ecosystem understanding — open models, infrastructure, safety, hosted vs in-house trade-offs
○ AWAITING REVIEW
Post-trained (SFT + RL) open models in-house. Frontier models in the pipeline.
I post-train open models — SFT and RL — for specific agentic use cases, so the experiences serve at scale and quality rather than paying frontier prices for every call.
The judge runs on an 8B open-weight VLM because it lives at the release gate and has to keep up with CI. We A/B'd a larger model: marginally better, significantly slower.
The generation path uses frontier models. That hosted-versus-in-house call, made on latency and cost rather than benchmark vanity, is the trade-off this bullet is asking about.
▸ TAP FOR EVIDENCE · +4% SYNC
Research prototypes to production systems, cross-functional and fast
○ AWAITING REVIEW
Every system here started unassigned. I own the pager after.
I scope it, build it, ship it, and own the pager afterwards. AgentX, the eval wall, the SDK, and Clauder all began as prototypes and are now load-bearing. I work best where the constraint is the problem itself rather than the process around it.
▸ TAP FOR EVIDENCE · +4% SYNC
Strong communicator across technical and non-technical stakeholders
○ AWAITING REVIEW
Conference talk, Confluent podcast, org-wide standards
The talk and the podcast are both in TRANSMISSION above — unedited, judge me on them. Internally I set the org-wide standards for AI tooling at Character.ai, which is mostly a persuasion problem rather than a technical one.
▸ TAP FOR EVIDENCE · +4% SYNC
Managing a small team of engineers
○ AWAITING REVIEW
Two orgs grown to 20+ engineers at Coinbase
Two orgs to 20+ engineers, company-wide technical strategy for Datastores, and a vendor replaced in-house at 90% cost savings. A small team is a smaller version of a problem I have run before.
▸ TAP FOR EVIDENCE · +4% SYNC
02 // THE PROOF
FIELD EVIDENCE // GAME SYSTEMS
0/6 REVIEWED
// SHIPPED INTERNAL WORK — AGENTS APPLIED TO INTERACTIVE AND GAME SYSTEMS.
CASE BOARD — FULLY PRODUCED MYSTERIES, GENERATED AND PLAYABLE IN SECONDS
EVIDENCE BOARD — AGENT-DRIVEN INTERROGATION, CONTRADICTIONS, TIMED ACCUSATION
The whole job description in one product.
Agents generate a coherent mystery, hold character through interrogation, and keep world state consistent across evidence and testimony — while staying solvable, under a timer, in front of players.
Every case is cast, filmed, scored and cleared by the eval wall before it reaches the board.
GENERATIVE NARRATIVEIN-CHARACTER AGENTSWORLD-STATE CONSISTENCYSOLVABILITY EVALSREAL-TIME PLAY LOOP
GENERATIVE NARRATIVE · IN-CHARACTER AGENTS · TIMED PLAY LOOP
PRODUCTION PIPELINE — SCRIPT GENERATION → AI CONTINUITY REVIEW → SEGMENT TTS → PUBLISH → PROMO VIDEO
The hard part is not generating one episode — it is generating episode seven and having it remember what happened in episodes one through six. StoryLab holds a series bible as durable state, runs arc review across the whole season, and gates each stage on a reviewer agent before anything reaches listeners. The series in this screenshot, Dungeon Hearts, is a LitRPG: HUDs, health bars, level bosses, permadeath. Game fiction produced by agents, shipped to a real audience.
LONG-HORIZON CONTINUITYSERIES-BIBLE STATEREVIEWER GATESMULTIMODAL: SCRIPT · TTS · VIDEOHUMAN-IN-THE-LOOP CONTROLS
LONG-HORIZON CONTINUITY · REVIEWER GATES · MULTIMODAL PRODUCTION
SHOT BOARD + THE CUT — 23 SHOTS, MULTIPLE TAKES PER BEAT, DRAG TO RE-ORDER, SUBMIT RE-CUT
Character identity has to survive 23 shots, and the timeline is the state — every re-cut re-derives what the agents must keep consistent. Takes are generated in parallel and scored before they enter the cut, so the human is making creative choices, not fixing broken frames. This is the same director's-chair model a games team needs when generated content has to be authored, reviewed, and shipped on a schedule.
SHOT-LEVEL AGENTSCHARACTER CONSISTENCYMULTI-TAKE GENERATIONHUMAN-STEERED CUTWEEKLY RELEASE CADENCE
SHOT-LEVEL AGENTS · CHARACTER CONSISTENCY · HUMAN-STEERED CUT
MAIN HALL — ROOM NAVIGATION, IN-CHARACTER SUSPECT DIALOGUE, LIVE EVIDENCE DISCOVERY
Nobody asked me to build this.
Every suspect is an agent with private knowledge, a motive to lie, and a story that must survive free-text interrogation — while the mystery stays solvable no matter what the player types.
World, agent architecture, front end, production ops: all mine. This is me unsupervised, for players not obligated to be kind.
HIDDEN-KNOWLEDGE AGENTSDECEPTION MODELINGSOLVABILITY GUARANTEESFREE-TEXT INTERROGATIONSHIPPED & LIVE
▶ PLAY IT YOURSELF — AISHEWROTE.COMHIDDEN-KNOWLEDGE AGENTS · DECEPTION MODELING · SOLVABILITY GUARANTEES
CLIP and LPIPS catch per-frame sharpness and caption match. They miss everything that decides whether a clip is usable: does it tell the story you meant, does it obey physics, does the character stay the same person, does the pacing land, is the audio synced. Judge Judy scores those axes per clip with a Qwen3-VL 8B model in about three seconds per 15-second clip — small on purpose, because it lives at the gate and has to keep up with CI. We A/B'd a bigger model: marginally better, significantly slower, not worth it. The judge is not prompted, it is trained: pairwise preference — is B a better story than A? — fine-tuned into the VLM with Bradley-Terry loss over real and deliberately corrupted footage, then calibrated against human annotations. Dataset curation through deployment through the retraining loop, in Python, owned end to end.
MULTI-AXIS SCORINGPAIRWISE PREFERENCE TRAININGBRADLEY-TERRY LOSSHUMAN CALIBRATION LOOPQWEN3-VL 8B · ~3s / CLIPLIVES AT THE RELEASE GATEOPEN SOURCE
MULTI-AXIS SCORING · HUMAN CALIBRATION · ~3s PER CLIP AT THE GATE
MIXED-MEDIA FEED — INTEREST TAGS, PER-POST REACTION SIGNALS, CREATOR AND AGENT CONTENT SIDE BY SIDE
The feed is the reason the eval work matters. Every reaction here is a live quality signal, and those signals are exactly what Judge Dredd correlates against offline eval scores to find which dimensions actually predict engagement. It is also the hardest surface to get right: mixed media, wildly different formats per post, generated and human content in the same grid, and moderation that has to hold at 20M+ MAU without flattening what makes the content worth posting.
MULTIMODAL FEEDENGAGEMENT AS EVAL SIGNALMODERATION AT SCALE20M+ MAU
MULTIMODAL FEED · ENGAGEMENT AS EVAL SIGNAL · MODERATION AT SCALE
03 // THE TOOLING
OPEN SOURCE // AGENT PRIMITIVES
58+ STARS · MIT
// SHARED TOOLING, BUILT IN THE OPEN — THE INFRASTRUCTURE LAYER UNDER THE AGENTS.
04 // THE CAREER
THE RECORD // MEMORY SEQUENCES
0/6 SYNCHRONIZED · +5% EACH
○ DESYNCHRONIZED
AgentX — The Autonomous Pipeline
Fully agentic multimodal pipeline at Character.ai: planner, executors for image, video, and audio, and reviewer agents gating every stage on trained eval scores.
30+ agentic turns, 6 specialized tools, no human in the loop after the prompt. Production-grade, observable, reliable at 20M+ MAU. Jeeves, an adaptive onboarding agent, runs on the same SDK.
MULTI-AGENT · GO · TRAINED EVALS▸ RELIVE · +5%
○ DESYNCHRONIZED
JudgeJudy & Judge Dredd
An open-sourced multimodal eval harness, then the self-improving layer that keeps it honest.
Judge Dredd correlates eval scores against production telemetry over MCP, finds which quality dimensions actually predict engagement, and feeds validated gates back into the runtime automatically. The judge underneath is a fine-tuned Qwen3-VL trained on pairwise preference with Bradley-Terry loss — dataset curation, training, calibration against human annotations, and the retraining loop, in Python.
PREFERENCE-TRAINED VLM · PYTHON · MCP▸ RELIVE · +5%
○ DESYNCHRONIZED
The Agent Primitives
claude-agent-sdk-go — the production SDK every agentic system at Character.ai runs on — plus Clauder, an MCP server for persistent agent memory.
Streaming, tool registries, hooks, MCP support, bounded goroutine pools. Clauder gives AI coding tools persistent memory and real-time inter-agent communication across sessions and repos.
GO · MCP · OPEN SOURCE (MIT)▸ RELIVE · +5%
○ DESYNCHRONIZED
The Datastores Campaign
Company-wide technical strategy for Datastores at Coinbase; grew Datastores and Observability to 20+ engineers.
Seven engines in production. Replaced an external observability vendor in-house at 90% cost savings. Architected Wallet-as-a-Service infrastructure processing billions in daily trading volume.
CLICKHOUSE · COCKROACHDB · TEAM OF 20+▸ RELIVE · +5%
○ DESYNCHRONIZED
The Streaming Platform Years
Self-Service Datastores Platform and the Data Abstractions team at Netflix, serving 200M+ members.
Saved engineering 10–20 dev hours per week across the org. Contributed to Netflix OSS (Hollow — delta-based in-memory sync). Entertainment infrastructure at a scale where a bad deploy is a headline.
OSS CONTRIBUTOR · 200M+ MEMBERS▸ RELIVE · +5%
○ DESYNCHRONIZED
Google, VMware & the First Systems
A Cloud Spanner optimization cutting latency up to 80%; before that, tech-led a 6-person team from zero to GA in 6 months.
Rebuilt GCP Networking Quota Management at millions of QPS. At VMware: datastores, stream processing, identity, and air-gapped enterprise readiness. Foundation laid as an officer in the IDF Intelligence Corps.
SPANNER · MILLIONS OF QPS▸ RELIVE · +5%
05 // ON THE RECORD
TRANSMISSION // ON THE RECORD
● 2 RECORDINGS RECOVERED
Why Your Video Judge Rewards Slop — and How to Fix It
A generated clip where the character freezes for four seconds can still score well — the judge rewarded gloss instead of what actually happened. Video breaks the text eval playbook: temporal consistency, shot continuity, and story coherence don't show up in a CLIP score, and a team watching clips on Friday doesn't scale.
The fix that stuck: stop scoring, start comparing. Pairwise preference — is B a better story than A? — trained into a Qwen3-VL judge with Bradley-Terry loss on real and deliberately broken footage, calibrated against human scores, running as a CI regression gate before any user sees a release.
PAIRWISE PREFERENCEQWEN3-VL JUDGEBRADLEY-TERRY LOSSCI REGRESSION GATE
// EVERY TEAM SHIPPING GENERATED CONTENT TO PLAYERS HITS THIS THE WEEK THEY GO LIVE.
PODCAST · INTERVIEWED BY CONFLUENT
Harnessing Data at Scale — Character.ai's Transition to WarpStream
The streaming backbone under everything above: 5 GB/sec of multimodal event data feeding the agents, the evals, and the telemetry loops. Confluent invited me on to walk through why we moved and what it cost us not to.
WARPSTREAM5 GB/SEC INGESTCOST-EFFICIENT STREAMING
▸ READ THE ENGINEERING BLOG POST06 // THE FILE
DATABASE // CV
OPERATIVE
BRIL, M.
Principal Engineer
SAN JOSE, CA
STATUS · ACTIVE
CLEARANCE · 20 YRS IN PRODUCTION
STATUS · ACTIVE
CLEARANCE · 20 YRS IN PRODUCTION
2024 — PRESENT
PRINCIPAL SOFTWARE ENGINEER
Character.ai
Technical and product strategy for agentic AI in generative entertainment at 20M+ MAU. AgentX, Jeeves, the agent SDK, and the eval wall gating every release.
2022 — 2024
SENIOR STAFF ENGINEER
Coinbase
Company-wide Datastores strategy; two orgs grown to 20+ engineers; 90% observability cost savings; billions in daily trading volume.
2021 — 2022
SENIOR SOFTWARE ENGINEER
Netflix
Self-Service Datastores Platform and the Data Abstractions team, 200M+ members. Netflix OSS contributor (Hollow).
2019 — 2021
SENIOR SOFTWARE ENGINEER
Google Cloud
Cloud Spanner optimization cutting latency up to 80%; GCP Networking Quota Management rebuilt at millions of QPS.
2015 — 2019
STAFF / TECH LEAD
VMware (acq. Arkin)
Tech-led a 6-person team from zero to GA in 6 months: datastores, stream processing, identity, air-gapped enterprise readiness.
2008 — 2012
BSc, EE & COMMUNICATIONS
HIT, Israel
Graduated while serving as an officer in the IDF Intelligence Corps.
// LEAP OF FAITH
INITIATE CONTACT
Everything in this archive is already running in production. The only open question is what I build next, and who I build it with.
Email me with the subject line FULL SYNC and I'll reply within 24 hours with a one-page plan for your first 90 days of agentic systems.
© 2026 MAOR BRIL · SAN JOSE, CA · MEMORY ARCHIVE v20.26
ALL SEQUENCES BATTLE-TESTED IN PRODUCTION
ALL SEQUENCES BATTLE-TESTED IN PRODUCTION