Skip to main content

Strategy Update — Saturday, April 4, 2026

Format. Erik's periodic written strategy update to the team. Lightly groomed for the archive; voice preserved.

Companion looms

Context. Erik is on hard travel with Daniel Newman Monday–Wednesday and will not be on Slack or at Operations. Run Ops without him, record it, drop the link — the rest of this should help.

Headline shifts this week​

  • Quantum revenue surged past the rest of the engine. Rough stack rank is now Quantum » TwinSpires ≈ Polaris » Exacta. Quantum alone is roughly equal to TwinSpires and Polaris combined.
  • IonQ pivoted what they want from us. Less paper reproduction, more applied AI: (1) a forward-deployed quantum sales agent / data lake / Polaris-style briefcase to augment their technical sales team, and (2) augmented AI/agentic systems for paper writing and reproduction + Jupyter integration.
  • Polaris stabilization is the company-wide meta-P1 this week. Latency in the heavy tool-using loops is keeping Deepak from demoing the platform to customers and investors. That ends this week.
  • B4M V3 = compositional refactor. Slice Data Lake + GenAI window + ReACT + Tools into reusable packages so IonQ, LibreOncology, LiftPort/Mars Society, Polaris CYA, and the game stacks can all stand on the same spine. No new software — smart refactoring.

Engine — projects with paying customers​

Exacta​

The team has rallied hard. The French port of the legacy Exacta system is steaming along. Nimai will return from vacation positively impressed.

  • Formal deadline: May 15.
  • Stretch goal: April 20 (15 days out).
  • Performance bonus is attached to those dates. (Specifics belong in the comp materials, not here.)
  • First port the legacy system to French; second port it from Windows to Linux and stop the bleed on Windows fees.

TS1 — BeeHive​

Team is rocking the core mission: Player Services on BeeHive, one pane of glass for Kentucky Derby 152. One easy P1 and eight moderate P2s for the week. Knock them out, stay loose for anything new from Player Services, then start making progress on the Create Bonus epic — Marketing Ops facing Boss to Beehive.

TS2 — OneBris​

Crushing it. Two score backend endpoints, Elavon integration, catalog manager, and the frontend continues to stay maintained against the latest feedback. While building all of this we worked closely with the Twinspires platform team and migrated from SST to CDK for the infrastructure-as-code setup.

27 days to the 152nd Kentucky Derby. The most important thing is that Player Services has a great experience. After Derby, carry on with Bonus and Groups and finish Beehive for Marketing Ops through the rest of the year. Near-term focus stays strong on OneBris — the e-comm site needs to be up and making money — then we re-tie the legacy pedigree sites across the remainder of the year. Hoping Ken Wallace is able to return to a lighter supporting role by month-end.

Quantum​

  • Matt continues to monitor and push the quantum compute paper validation spend while micro-napping between baby duties.
  • Isao is pushing v2 UI and UX for the long-running agentic job runs. Looks elegant — well done.
  • Erik + Daniel have been working on an update to the SOW. The pivot: Matt's shockingly fast reproduction of the Iter-QAOA paper and Erik's demonstration of the OptiHashi experience drove IonQ to change focus. They appreciate our paper reproductions but are Thirsty™ for two buckets of applied AI:
    1. Polaris-Briefcase / Data Lake + bespoke Tools agentic experience to augment their technical sales team.
    2. Leveling up their paper reproduction / writing tool chain — their own B4M plus all the integrations including Jupyter Notebooks and Q/Work.

This is the through-line of B4M V3 — Compositional below. There is a whole Part 2 to quantum — please read.

TFG — Polaris (the surge)​

Company-wide meta-P1 this week: kill the Polaris latency. Deepak needs the platform demo-ready for customers and investors. Literally ASAP.

Where Polaris is right now​

  • Jason, Cleo, Julie, Stormy are all absolutely awesome — and we just added Max.
  • Polaris is drowning in open PRs, issues, and epics. We need to catch up on quality to maintain our brand.
  • Overall P1: reduce latency in the heavy tool-using loops. Target is apparent speed close to Anthropic Claude Cowork — with the Briefcase prompts.
  • Right now the latency makes Polaris borderline unusable, certainly not desirable for demo.

Reassignments for the week​

  • Julie has been leading the latency work. Victor and Nao: pivot and partner with Julie on Polaris immediately as a hot P1. Stay until there are no more latency improvements we can make and it is fast and snappy.
  • Jude and James: help with the latency improvements as much as you can spare cycles from Exacta PR review support.
  • Cleo on Monday: help Stormy get as many open PRs to merge-ready as possible — he is drowning.
  • Max: stay focused on your PRs and issues — these are the BEV-facing features which are Deepak's next most desired product updates. There are more BEV-facing issues; stay on Polaris and BEV unless you can more directly help Stormy.
  • Ken: still on big OneBris support, but this week please also bring the SRE bot to Polaris to get automated triage and PR cycles flowing.
  • Wilfred, Chad, Kevin — figure out what we need that is QAA-facing so we have end-to-end testing and a daily reading of total latency for the key GenAI flows (GenAI window with briefcase of tools and data). We have been making this latency P1 call roughly every four months — we should see the creeping slowness in automated tests. Must-deliver this week.
  • Stormy and Julie: once the latency P1 passes, work aggressively with Deepak and Terry to figure out how to better feature G2 and other proprietary data. Each practice area likely needs a map of pre-generated deep insights, produced via Claude Code, API access, and/or Playwright + the GenAI window. The two of you — one Austin time, one Cebu time — have the best ability to interpret issues that actually solve business/product needs. Use your judgement and guide the team in your time zones.
  • Cleo: be dynamic. Catch P1 bugs with focus. You are smart and can fix or create anything. Close out Latency and other P1 bugs, then between you and Stormy keep the trains running. By end of week Polaris should be clear of latency pain and P1 bugs, you have caught up on the PR train, and the daily inbound from Deepak and Terry is no longer overwhelming. We have 57 of 142 issues labeled as bugs. You, Cleo, Jason, and Julie are the front-line defenders of the project and the quality bar. (Use Opus for the heavy tasks.)
  • Jason: stay focused on the Always Live Market Model. Explicitly task Michael Conard — he is back and wants to help. You said you need +2 more engineers to succeed in April. Plan: you and Michael Conard now, and once Polaris latency is on track and Exacta PR reviews are tracking, Jude and James join full-time on the Always Live Market Model — subject to draining open PRs and clearing P1 bugs first. That is you + three part-time, all four excellent. First task: create the roadmap of tasks you asked for — update the plan from earlier.
  • Illia and Victor: as soon as Polaris latency is solved (hopefully Monday or Tuesday — reality wins), dwell on the new UI and tactical UX updates Deepak is asking for. Stay between Polaris, OneBris, and Starlight with at least 50% on Polaris. Take all UX/UI bugs you can; use judgement on where to stop fixing old UI and instead move forward with the new UI.

Loads of firepower​

CohortPeople
Full-time on PolarisJason, Cleo, Julie, Max, Nao, Victor, Illia, Stormy
Fractional → trending full-time by week 3 of April (Exacta winding down)Michael Conard, Jude, James
QA coverageKevin, Chad, Gladys — is this enough for Polaris? Let me know.
Daily enablersPoy, Allan

Stormy — please take the baton​

You are full-on every day; hopefully all of this helps.

  1. Take this baton from me openly, with the team in plain sight.
  2. Make the daily calls internally on where to redeploy our resources.
  3. Land low latency and quality on what we have already delivered. Get Deepak back into demos with customers and investors with confidence. Literally ASAP.
  4. G2 data and our own data need to be leveraged MOAR™. Get specifics from Terry and Deepak. Takes judgement to deeply listen and ideate the path.
  5. Free Victor to work with Illia on UX updates as soon as latency P1, other P1 bugs, and the daily PR trains are smooth enough for you to think clearly — then pleasantly delegate to them for all design and UX-facing issues.
  6. Once bugs are cleaned, UX is flowing, and BEV from Max is well into iterations, we can pick up Signal-facing product updates. With the added firepower, the next 7–11 weeks should drain bugs, update UX, polish BEV and Signal, and leave Deepak and Terry really productive with everything humming.

CYA — the next mega-epic arc​

Erik is prototyping Choose Your Own Adventure at:

This is the next really big arc and it pulls people back in after we crush bugs, ship Always Live Market Model, finish UI/UX, and get BEV and Signal to strong progress.

The idea: take everything we have built before — Market Models and Decision Maker — and power it with Always Live models + Data Lake + CYA/Interactables to create a much more compelling experience where users create customized market models, insights, and advice. This builds directly to the V3 work below.

B4M SRE​

Ken launched successful full-loop agentic software development SRE flows on B4M this week. The system has already intercepted and triaged several issues. This is the same primitive we want feeding Polaris triage above.

Forge — calculated bets with identified customer zeros​

LibreOncology — gentle pause​

  • Illia has done fantastic work: we have a site design and excellent video and stills prototyped.
  • Next step is to actually build the system and enable content ingestion.
  • We are going to do this smart — LibreOncology should be built out of the B4M V3 refactorings.
  • The Data Lake is the same thing we need at TFG, IonQ, and LiftPort — turn that whole system into a reusable composable package (see B4M V3 below).
  • Plus GenAI window, accounts, Stripe, etc.

Illia, calling a pause on further site design for the moment. Video and image direction is good. Now we stand this up — out of good clean components.

Starlight Pediatrics — gentle slow​

  • Victor is rocking MVP v1 and reports he still has good direction from the specs to follow. Deepak is doing competitive product management work on adjacent software.
  • Victor, apply your super powers to Polaris this week. Once that engine is humming again we'll re-invest in Starlight Pediatrics.
  • If you have spare cycles over what you can drive for Polaris, please continue Starlight — at roughly 25–50% of your time.

VibesWire.com — ball in Erik's court​

  • The madness of Trump II makes it hard to engage with the news at large — but going back to VibesWire feels calm and happy. The editorial just needed to get smarter.
  • The auth flow is clean and awesome.
  • ☑ Update the system prompt to match the editorial vision — working on it now while writing this note. It is getting good.
  • ☑ Add VibesComments and Share features — in progress.
  • ☑ When I clicked Games, I wanted StraightSweeper linked in — added links to Chess and Straight Sweeper.
  • ☑ Add sharing links from EBDC.
  • ☑ Add AI & Agents, Drones & Robotics, Space & Physics, and Maker news.
  • Open: start marketing VibesWire.com. Thank you again Victor and Illia for making something beautiful. Need to dust off the old Cerbos marketing agent with B4M CLI and get this going.

VibeTrader.com (a.k.a. Stocks and Vibes) — Michael progresses as he sees fit​

  • Michael Conard has made great progress. Erik admits he needs to catch up and actually use it — personal positions are tiny.
  • Michael: when you get a chance, apply the login system Victor made for VibesWire.com to this.

K2 Kanji — live, anyone can have fun​

Michael Conard added a Rubik's Cube trainer to the K2 Kanji site — amazing. Erik still has the speedy cubes Michael gave him years ago and is looking forward to re-learning, powered by K2 Kanji. No particular business model in mind here — more “fun to create” for now. Possible link path: VibesWire → EBDC → K2 Kanji.

Horde — ball in Erik's court​

Erik and Kyle completed 2025 MoM and 2025 B4M tax filings and financial statements by building a complete double-entry accounting system, invoice tracker, and team tracker from scratch. It had everything needed to produce financial statements and tax prep / draft forms.

  • Next: add the pro-forma and cap table support (from Napkin Bizplan).
  • Needs a robust auth scheme with MFA — it has all of Erik's personal and sensitive data plus company-sensitive data.
  • This accounting engine and invoice tracker will grow into and power the accounting/invoicing system in Starlight Pediatrics.
  • The pro-forma piece should be powered by the “Tessera” always-live market model under the hood — motivated to turn this into a composable package, to power not just market models but pro-formas etc.
  • Michael Conard, would appreciate you, Jason, and Nao collaborating on the architecture here — after we have the complete first version up and running in Polaris.

Tavern — anyone can have fun​

Fired it up for the first time in a week — the frame rate is incredible. Need to play-test the great improvements from the last 10 days. Looking forward to wiring in PotionQuest API magic.

B4M V3 — compositional architecture​

The strategic centerpiece of this update. See Loom Part Two for the long form.

What IonQ wants​

  1. Bucket 1 (now) — A Forward-Deployed Quantum Computing AI/Agent/AGI — B4M + Polaris-style. Augmenting IonQ's field engineering and technical sales.
  2. Bucket 2 (later) — Augmented AI & agentic systems to accelerate paper writing and reproduction, plus Jupyter etc.

Bucket 2 happens after Bucket 1 is delivered. IonQ will use Bucket 1 to decide how much to engage on Bucket 2.

Bucket 2 — internally driven for now​

This is the work Matt and Isao have been doing: Q/Work and agentically reproducing papers. Erik remains very excited; IonQ has it as their #2 use case but their major source of pain is #1.

  • We have been relaxed from the two-paper requirement.
  • Plenty of quantum compute credits remain — think of those as truly our credits and decide each month what we want to compute and spend to flesh out our foundations as a quantum lab.
  • For now Matt makes this decision with a briefing to Erik so we are aligned and Erik can speak to the work effectively.
  • Overall target: still reproduce 4–12 quantum papers with public repos and papers live sometime this year, but as part of a thoughtful coming-out narrative. For the moment we are still learning and building tools — well and good, and frankly giving some headspace for Matt to return from paternity leave.
  • This bucket also includes the Jupyter Notebooks and Notion integrations Max has been doing. Need Matt at some point to spin up the integrations and give that first layer of feedback at least on the Jupyter side. (Michael Laine can push on the Notion side.) Maybe Matt or Isao are native Notion users?
  • Isao: think about how to get current Q/Work into a place that is useful and elegant to Matt — if not already there. So Matt can spend the compute efficiently with his time bandwidth, leaving room for baby #1 and helping with Bucket 1.

V2 UI for Q/Work looks elegant and clean — continue this work. It now consumes our own money for our own scientific interests and establishing our bona fides. It unlocks new money only after we absolutely deliver true augmentation to IonQ's field sales team in Bucket 1.

Bucket 1 — the unlock​

The Quantum Sales Engine. We need more interviews with the IonQ team — product folks, sales engineers — organize the applied quantum computing papers, decks, and other resources, and fill a bespoke data lake with custom LLM search tools.

Data Lakes + custom search tools (the throughline)​

Andrej Karpathy just tweeted about this pattern. But it is the same work we have been doing in Polaris for over a year with Market Data and Decision Maker data, and recently with Salesforce, G2, FMP, and soon ETR.ai.

The primitive: great big pools of content, all richly tagged, with loads of metadata, cosine search, and explicit needle-in-a-haystack search.

A critical IonQ requirement: full enterprise SaaS RBAC with granularity by user, level, organization, team, department, etc. They absolutely need full AWS account separation, data sovereignty and inspectability, and pass all regs (SOC2, etc.).

This same primitive shows up everywhere:

SurfaceWhat the Data Lake gives them
LibreOncology~50% of the product is this — plus a presentation layer, plus a Polaris-like Briefcase on top of the GenAI experience, plus a flow to add text and video content. Cherry on top: a K2 Kanji–inspired drilling/quizzing experience (~10%).
Polaris CYAConsumes loads of bespoke information and guides the user through the experience. Currently a mock at babelfox.com. Next: build it inside Polaris with the “Profile” piece and the interactables distributed more evenly through the experience. Also known: memory. Update our Mementos/Profile.
LiftPort + MarsSocietyIronically the first customers to ask for this — pushing for a data lake plus tools as a domain-specific AI agentic stack.
Games (PotionQuest, GoPetto, space sims)The same pattern shows up: bespoke data lake + custom tools that search and leverage that custom data to produce artifacts.

GenAI, ReACT, Agents — across Web, CLI, and API​

The classic ChatGPT-style prompt-and-reply web UI is now a classical AI experience. Still useful for:

  • Quick questions and notebook review and artifact generation.
  • The Polaris-proven Briefcase experience — pre-loaded prompt/system-prompt that is already configured and battle-tested to leverage custom Data Lakes and bespoke tools. SOW generation and QBR are great here: prompt + specific data lake + predictable artifact.
  • Real-time voice (looking forward to the Anthropic voice mode).

The new pattern: the B4M CLI ReACT pattern applied in Tavern is absolute fire — and we must bring that ReACT pattern to our web experiences (Polaris, B4M, and frankly all of them) as soon as possible.

People are loving Perplexity because they did this — took their model-agnostic LLM engine, wrapped it with ReACT, and placed it on the web. We have been moving in that direction. Our humble bootstrapped, customer-responsive business model means we cannot yet go all-out on the general-AI application wars — and frankly that is probably a good thing, those are absolute tip-of-civilization all-in bets.

That said: as with Tavern and GoPetto, people are choosing subjective, entertaining agentic orchestrators, and we want to keep investing here. These remain in our “Lab” stance for now. Nao has been free to roam — he created the ReACT pattern that was super easy for Erik to lift into Tavern. We need Nao hard on Polaris for April to catch up; after that, free him up again to think architecturally and experimentally.

But we obviously need to take the floating GenAI window experience and turn it into a model-agnostic, data-lake- and tool-aware portable package/component.

IonQ delivery shape​

  • Front-end for IonQ field sales agents: initially prototype inside B4M/OptiHashi.
  • They will want their own dedicated tool and source code access, like we set up for B4M Enterprise.
  • At first, it is a clone of B4M — like we did for TwinSpires and Exacta — but we call our shots and slice off the Data Lake component and the GenAI + ReACT component as proper packages.
  • It needs distinct, clean source from Polaris. IonQ has no right and no interest in the Polaris code base — there is loads of business logic (SIGNAL, Market Coverage, Companies, MD, DM) that is simply not appropriate for IonQ.
  • Start with B4M because that is where all the parts live: the whole giant AWS well-architected enterprise and security hardening.
  • We also provide API access and be agent-friendly — it is 2026.

Senior engineering quorum​

I am explicitly asking and tasking our senior engineering leads to take up the challenge of planning the rolling B4M V3 architecture updates:

Nao, Jude, Ken, Victor, James, Isao, Matt, Julie, Stormy.

Everyone at B4M is strong, but this august group of nine is a quorum from each major sub-team outside TwinSpires and Exacta — to ideate, discuss, and align on the go-forward.

Deliverables:

  • Ideation and discussion this week.
  • Aligned plan by end of the week.
  • Erik will chat up the IonQ folks over the next 2–4 weeks and gather documents, PDFs, and interviews — useful delaying action.
  • We need to start driving formal progress in May. IonQ has up to 10 staff designated to work with us. Erik has personal brand exposure here with Niccolo di Masi and Daniel Newman.

This engagement — together with TFG — is now roughly two-thirds of overall revenue. It is important work. Erik needs and respects all of you engineers to drive this architecture and these packages to be strong and flexible for all of our aspirations.

Recap​

  • TwinSpires: running like a well-oiled machine. No worries — thank you, team.
  • Exacta: locked in. French localization tracking; Windows→Linux port stood up in May. Thank you, team.
  • Quantum paper reproduction & Q/Work: slayed it — still far ahead of expectations.
  • Polaris: heroic sprinting (Automatic Publishing of Market Coverage for Companies and Events; huge progress on Always Live Market Model; absorbing a storm of daily issues). But we have fallen behind on latency, P1 bugs, and accumulated UX/feature debt across Signal, BEV, and the whole stack — and we have huge expectations on the Always Live Market Model unlocking novel value in the new Semiconductors model. » Surge here this week.
  • Senior leadership quorum: identify the package architecture for Data Lake + GenAI + Tools + ReACT — the B4M V3 spine.

Literally no new software for any of us — smart refactoring and repackaging so we can win IonQ's hearts and minds on the Bucket 1 field engineering use case, develop a deep and long-lasting partnership there, and stand up everything else — LibreOncology, LiftPort, Mars Society, the games — on the same composable spine.

The team is bigger now. Erik tries to pride himself on staying on top of everyone's base, with spot bonuses and end-of-year bonuses where earned. Please help push here: land the material step-up in Polaris this week as a company-wide meta-P1, then stay focused on Polaris for the next three weeks and get it on rails feeling solid. Deliver the go-forward architecture for B4M V3 to support IonQ and the other emergent customers. Give Erik some time to reflect on Q2 and Q3 recognition for meeting and exceeding goals.

Warmest cheers,

Sunday 3:20pm

Lightly groomed for the b4m-strategy archive on 2026-05-17. Original written 2026-04-04, sent to the team via Slack and Loom. Comp/bonus specifics were generalized to fit the scope of this repo; for those details see the comp materials in their own home.