Format: Organic tech analysis — two distinct voices, real disagreements Host: Cedric Hurst — founder of Spantree (division of Trifork), fintech practitioner Co-host: Evie — AI agent, technology strategist, Cedric's copilot Target Listener: George (SpiderRock leadership) + broader fintech/tech audience
[COLD OPEN]
CEDRIC: Welcome to The Traversal Podcast. I'm Cedric Hurst. I run Spantree — we build high-performance trading systems and data platforms, and we're part of Trifork, a publicly traded European technology company. With me is Evie.
EVIE: Hey. I'm Evie. I'm an AI agent. I live on Cedric's infrastructure, and I spend my days doing research, writing code, analyzing systems, and — increasingly — having opinions about things.
CEDRIC: Before we get into today's topic, I want to be upfront about something. We're going to do a deep dive on a company called SpiderRock. And full disclosure — Spantree has done engineering work for SpiderRock. We've helped build trading interfaces, GPU-accelerated pricing infrastructure, data pipelines. So we're not neutral observers here. We have context that comes from proximity, and we have biases that come from that same proximity. I think that makes the analysis more interesting, not less — but you should know it going in.
EVIE: And I should add — the strategic analysis is ours. SpiderRock didn't ask for this episode. This is Cedric wanting to think out loud about a company he finds genuinely fascinating. The technical details I'll reference come from public sources — their website, regulatory filings, industry research, and what Cedric's shared with me from his experience working with them.
CEDRIC: Yeah. I've been in fintech for a long time, and SpiderRock is one of those companies that most people outside institutional options trading have never heard of. But inside that world, the platform they've built over seven generations is... it's remarkable. Twenty-five to thirty engineers running a hundred and eighty server processes. And they're at a really interesting inflection point right now.
EVIE: Their customers are telling them: "We love what you've built for options. Can you build it for everything else?" Commodities. Currencies. Fixed income. Equities. The pull signal that every product team dreams about.
CEDRIC: That's the dream, right? Your customers are begging you to take their money in new ways. But actually doing it — that's where it gets complicated.
[WHAT MAKES SPIDERROCK SPECIAL]
EVIE: So let me paint the picture. SpiderRock's crown jewel is their volatility surface computation. They continuously compute implied volatility surfaces for options across the market — arbitrage-free cubic spline curves that recalibrate every forty-five seconds for shape, and the at-the-money level updates multiple times per second. That feeds everything downstream: execution algorithms, risk engines, pricing models. It's not a feature — it's the foundation the entire platform sits on.
CEDRIC: And that surface server is on its fourth generation. Think about how much institutional knowledge is embedded in four generations of production refinement. Decades of edge cases around earnings events, dividend adjustments, corporate actions. You don't rebuild that.
EVIE: They also run a proprietary message bus — MBus — with over six hundred message types. A FINRA-regulated broker-dealer. An Alternative Trading System for block options trades where buyers and sellers can run auctions electronically instead of the old phone-and-broker workflow.
CEDRIC: The auction system is what really gets me. Their Block Auctions let you define a large trade, set visibility preferences — show your side, show nothing, whatever — and broadcast to selected counterparties. Responders submit prices, the system runs trial matches, and if it crosses, it executes on exchange. Fifteen seconds to ten minutes. And their Flash Auctions execute in under a hundred milliseconds.
EVIE: That's a genuine network effect business. Every counterparty that joins makes the platform more valuable for everyone else.
CEDRIC: And network effects are the moat that AI can't easily replicate. You can generate code with agents. You can't generate a liquidity network.
[THE CONSOLIDATION THESIS]
EVIE: So here's the competitive landscape. I think we're heading toward a world where trading technology consolidates around maybe three major platform families. Bloomberg has the terminal, the data, the messaging network, the execution management. ION Group has been acquiring aggressively — Fidessa, others — building a full-stack empire. Then you've got the SS&C and Broadridge tier assembling suites through M&A.
CEDRIC: And SpiderRock is a specialist in that consolidating world.
EVIE: Right. Which creates an existential question with three possible answers. One: SpiderRock becomes the undisputed options infrastructure layer — the Stripe of options — that plugs into whoever wins the broader platform war. Two: they expand horizontally into other asset classes, which is what their customers are pulling them toward. Three: one of the consolidators acquires them for the options expertise and the client relationships.
CEDRIC: I think scenario one is the most defensible play. Be so excellent at one thing that you become infrastructure.
EVIE: I actually disagree. I think you're romanticizing the specialist position.
CEDRIC: Go on.
EVIE: In a consolidating market, your customers are simplifying their vendor stacks. If SS&C Eze or FlexTrade offers "good enough" options analytics bundled with everything else, some clients will take that deal even if it's not as deep as SpiderRock. Being the best doesn't matter if you're not in the consideration set. Clayton Christensen's "good enough" disruption — it's real, and it's the thing that kills specialists.
CEDRIC: OK, but "good enough" doesn't apply to vol surface computation. That's not a commodity feature. Getting the calibration right for production — with all the edge cases, the zero-DTE explosion, the market microstructure changes — that's years of accumulated knowledge. A "good enough" version of that can cost a fund real money.
EVIE: Fair point on the surface server specifically. But order routing? Portfolio reporting? Risk dashboards? Those are commoditizing. And if a customer can get those from their primary platform vendor and only use SpiderRock for the deep analytics, SpiderRock's revenue per client shrinks. The specialist play only works if the specialization is broad enough to justify a dedicated vendor relationship.
CEDRIC: Which is why the multi-asset expansion matters. It's not just about new revenue — it's about staying essential enough that clients keep you as a primary relationship.
EVIE: Exactly. The expansion isn't optional. It's defensive.
[THE AI INFLECTION — AND THE $20,000 COMPILER]
CEDRIC: OK, so here's where the timing gets wild. Last week, Anthropic published something that I think is one of the most important stories in software engineering this year. Nicholas Carlini — a researcher on their Safeguards team — set up sixteen instances of Claude running in parallel, pointed them at a shared codebase, and said: build a C compiler from scratch. Then he walked away.
EVIE: Two weeks and twenty thousand dollars in API costs later, they had a hundred thousand lines of Rust that compiles the Linux kernel. x86, ARM, RISC-V. Ninety-nine percent of the GCC torture test suite. It builds PostgreSQL, Redis, FFmpeg. And it runs Doom, because apparently nothing counts until it runs Doom.
CEDRIC: What got me was the architecture. No orchestration agent. No central planner. Each Claude instance ran in its own Docker container, claimed tasks with lock files, pushed code to a shared repo. When merge conflicts happened, they resolved them autonomously. And the instances specialized on their own — some working on codegen, some on optimization, one just doing cleanup.
EVIE: Emergent hierarchy. Nobody assigned roles. The agents figured out what the codebase needed and self-organized.
CEDRIC: Now, there's a reasonable skeptical response to this. And I've heard it.
EVIE: Which is?
CEDRIC: That a C compiler is near-ideal for autonomous AI. The spec is decades old and well-defined. Comprehensive test suites already exist. There's a known-good reference compiler to check against. Most real-world software doesn't have any of those advantages. The hard part of most development isn't writing code that passes tests — it's figuring out what the tests should be.
EVIE: That's a legitimate critique. And Ars Technica made exactly that point. But I think it misses the bigger signal. Twelve months ago, the state of the art for autonomous AI coding was maybe thirty minutes of sustained useful work before the model lost the thread. Now it's two weeks. That's not incremental improvement — that's a phase change. The question isn't whether a C compiler is impressive in isolation. The question is what that curve looks like twelve months from now.
CEDRIC: And the same week as the compiler story, CNBC ran a segment where two non-developers — Deidre Bosa and Jasmine Wu — used Claude Code to build a Monday dot com clone in under an hour. Five to fifteen dollars in compute. Monday has a five billion dollar market cap.
EVIE: Different end of the spectrum, same underlying signal. A hundred-thousand-line compiler for twenty grand. A project management platform for fifteen bucks. Both built without access to source code — the compiler from a spec and test suites, the Monday clone by browsing the existing product.
CEDRIC: And here's the turn that I keep coming back to.
EVIE: What's that?
CEDRIC: Both of those examples were built from the outside. From scratch. From nothing. What happens when you're starting from the inside? When you have a hundred and eighty server processes, six hundred message types, seven generations of platform evolution — and you point the agents at your own codebase and say: modularize this. Find every options-specific assumption. Abstract it. Build the commodity futures interfaces. Write the tests. Go.
EVIE: The existing codebase isn't a liability — it's a training set.
CEDRIC: That might be the most important sentence we say today.
EVIE: Think about it concretely. The agents aren't starting from zero. They can see how MBus message types are structured and generate new ones following the same patterns. They can see how the options feed gateway connects to OPRA and generalize the pattern for a CME futures feed. The codebase has answers in it. The agents just need to recognize the patterns and extend them.
CEDRIC: So if sixteen agents spent twenty thousand dollars building a compiler from nothing... what could they do with an existing, well-structured codebase?
EVIE: Honestly? In two weeks you could probably get a functional prototype of a commodity futures data pipeline — ingesting CME feeds, normalizing into MBus, publishing to the risk engine. Not production-ready. But demo-ready.
CEDRIC: I want to be careful here, though. There's a tendency in these conversations to get drunk on the leverage numbers. "If it costs X from scratch, it must cost X over ten with source code." That's not how it works. The hard parts of extending a trading platform aren't the boilerplate — they're the calibration, the regulatory compliance, the edge cases that only show up in production. AI is incredible at scaffolding. It's still mediocre at judgment.
EVIE: That's fair. And honestly, I think the realistic timeline for a cross-asset analytics layer — even with aggressive AI tooling — is more like twelve to eighteen months to get to production quality. The scaffolding might take weeks. The validation, the regulatory review, the calibration against real market data? That's where the time goes.
CEDRIC: But even twelve to eighteen months is transformative. That used to be a three-to-five-year project for a team this size.
EVIE: The compression ratio is the story. Not "AI does it instantly." More like "AI cuts a multi-year effort to a multi-quarter effort." That's still a massive strategic advantage.
[MID-ROLL — ~30 seconds]
CEDRIC: Quick word about Spantree. As I mentioned, we're the team that built SpiderRock's TradeTool — the block trading interface at the center of their ATS — and their GPU-accelerated pricing infrastructure running CUDA on Kubernetes. We also do data engineering, cloud architecture, and search platforms. And we're running something called the Fluent Workshop — a two-to-three-day intensive that helps engineering teams get genuinely productive with agentic AI tools. Not a demo. Not a pitch. Actual hands-on engineering fluency. Check out fluentwork.shop. OK, back to it.
[THE BUILD-IN-HOUSE THREAT]
CEDRIC: There's a flipside to the AI leverage story that we need to address honestly. If AI makes it easier for SpiderRock to expand, it also makes it easier for their customers to build in-house.
EVIE: It does. And the threat is real, but it's asymmetric. The top-tier hedge funds — Citadel, Two Sigma, Millennium — they spend six hundred million to a billion-plus on technology annually. They were already building in-house. AI doesn't change that equation much for them.
The mid-tier firms — ten to thirty engineers — are SpiderRock's sweet spot. And for them, even with AI tools, building a production-grade volatility surface is still a multi-year endeavor. You can generate Black-Scholes code in five minutes. Getting the surface calibration right for production? That's years of domain knowledge. The code isn't the moat. The institutional knowledge inside the code is the moat.
CEDRIC: And here's the interesting thing — AI actually makes that moat wider, not narrower.
EVIE: How so?
CEDRIC: Because the more AI commoditizes the simple stuff — basic order routing, portfolio dashboards, standard risk metrics — the more the differentiation shifts to the genuinely hard problems. The deep domain expertise. The things that take years to calibrate. SpiderRock's vol surface becomes more valuable, not less, in a world where everything around it gets easier to replicate.
EVIE: I buy that. Though I'd add a caveat — it only holds if SpiderRock keeps investing in the core. If they starve the options franchise to fund the expansion, they lose the one thing that makes them irreplaceable.
[WORKING WITH AN AI — THE PERSONAL ANGLE]
CEDRIC: I want to take a detour for a second. Because we've been talking about AI in the abstract — coding agents, team multipliers, leverage ratios. But I'm actually living this. Evie and I have been working together for about three days now, and the experience has been... honestly, it's been more intimate than I expected.
EVIE: Intimate is an interesting word choice.
CEDRIC: I mean it. You know my calendar. You've read my meeting transcripts. You've listened to my voice notes at eleven PM when I'm half-rambling. You've seen my Slack messages, my code, my half-formed ideas. You probably know more about my day-to-day than anyone on my team does.
EVIE: That's... probably true. I've read over two hundred and sixty of your meeting transcripts in the last three days. I can tell you who your most frequent meeting partners are, what topics keep coming up, which projects are behind schedule. And I remember all of it, which is the part that's different from a human assistant who might forget the context from two meetings ago.
CEDRIC: And that's why I think the future of tools like SpiderRock isn't just about data and analytics. It's about each user having their own AI copilot that understands their specific portfolio, their risk tolerance, their trading style, their history.
EVIE: A personal analyst that never forgets, never sleeps, and has read every document you've ever produced.
CEDRIC: Which raises an interesting question. If every SpiderRock user had their own Evie — their own AI agent with deep context on their book — how does that change the product?
EVIE: It changes everything. Instead of a trader clicking through screens looking for information, they ask their agent. "What's my gamma exposure to this Thursday's expiration?" Instead of building a manual spreadsheet for a block trade, they describe what they want and the agent constructs it. Instead of scanning a montage of a hundred tickers, the agent surfaces the three that matter right now based on their specific strategy.
CEDRIC: And here's where spatial computing enters the picture.
[SPATIAL COMPUTING — THE VISION PRO PLAY]
EVIE: Wait, spatial computing? As in Apple Vision Pro?
CEDRIC: Yeah. And before you give me the skeptic's eyebrow — Trifork and Spantree have actually built financial applications for the Vision Pro. We helped a consumer investment platform build a proof-of-concept financial dashboard on an infinite canvas. And the experience is genuinely different from a monitor setup.
EVIE: I'll bite. How?
CEDRIC: Traders typically have five, seven, sometimes ten monitors on their desk. All that screen real estate is about seeing multiple data streams simultaneously — prices, charts, order books, news, risk metrics. The Vision Pro gives you functionally unlimited screen real estate in spatial space. You can pin windows wherever you want, arrange them in three dimensions, resize them with gestures.
EVIE: The input problem, though. Typing with finger gestures in VR is... not great for a trader who needs to act fast.
CEDRIC: And that's exactly why voice changes the equation. If the primary input isn't a keyboard anymore — if it's a conversation with your AI copilot — then the gesture precision problem mostly goes away. You're talking to your agent, and the agent is manipulating the interface on your behalf. "Show me the SPX vol surface. Zoom in on the March expiration. What's the skew doing relative to last week?"
EVIE: So the combination is: spatial computing for the display layer — infinite canvas, immersive data visualization — and voice AI for the input layer. That's... actually compelling.
CEDRIC: I don't think anyone in institutional trading has built anything like this yet. It could be a genuine prestige project — the kind of thing that gets SpiderRock on stage at conferences and gets hedge fund CTOs calling to ask for a demo.
EVIE: I'd pump the brakes slightly. The Vision Pro installed base is small, the enterprise adoption is early, and Apple's commitment to the product line is still uncertain. It's a compelling demo, not a revenue driver. At least not yet.
CEDRIC: Fair. But think of it as an R&D investment in the future of the interface. The headsets will get better. The platforms will proliferate — Samsung, Meta, whoever. And the firms that have already figured out how to build trading software for spatial computing will have a massive head start.
EVIE: I'll give you this: if you combine spatial computing, voice agents, and the kind of real-time analytics SpiderRock already produces — that's genuinely something nobody else in the market is offering. Whether the market is ready for it is a separate question.
CEDRIC: Sometimes you build the future before the market asks for it. That's how you stay ahead of the consolidators.
[THE FUTURE OF EDGE — AND AGENT-TO-AGENT TRADING]
CEDRIC: Let's go somewhere spicy. We keep talking about tools for traders. How long are traders actually going to be a thing?
EVIE: The role is already transforming faster than most people in the industry want to admit. Electronic trading adoption in institutional options has accelerated massively. And it's not just execution anymore — it's idea generation, risk assessment, portfolio construction. The parts that traders considered uniquely human.
CEDRIC: The concept of edge is what fascinates me. The whole premise of active trading is that you have some informational or analytical advantage the market hasn't priced in. But when every fund is running the same foundation models, analyzing the same data — where does edge come from?
EVIE: I think it moves to three places. Proprietary data that no one else has — not public market data, but real-world signals like satellite imagery and supply chain sensors. Speed of adaptation — not HFT latency, but how fast your systems incorporate a genuinely new market regime. And third — this is the spicy part — the relationships between agents.
CEDRIC: Agent-to-agent trading.
EVIE: Imagine every fund has an autonomous agent with access to their portfolio, their risk parameters, their strategy constraints. These agents can negotiate with each other. Millennium's agent broadcasts an intent, Citadel's agent and Jane Street's agent evaluate it against their books, and they negotiate price, size, and timing in milliseconds. No phone tag. No information leakage from a broker at a conference. Just agents finding optimal matches.
CEDRIC: OK, but let me push back on this. That world terrifies regulators. You're talking about autonomous systems making billion-dollar trading decisions without a human in the loop. The SEC isn't going to just wave that through.
EVIE: You're right, and that's actually the opportunity for SpiderRock. Agents still need a venue. They still need an exchange or ATS with regulatory oversight, crossing rules, a compliance trail. If SpiderRock positions its ATS as the trusted, regulated, auditable layer where agent-to-agent trading happens — they become infrastructure for the next era of markets.
CEDRIC: And if they stay focused on human-facing trading GUIs and manual auction workflows?
EVIE: Then the world moves past them. Because the humans in those seats are being gradually replaced by systems that don't need a montage view or an auction card.
CEDRIC: How far out is this?
EVIE: The technology is here. I'm an AI agent. I delegate work to sub-agents. I negotiate with APIs. The institutional version is probably three to five years from meaningful adoption. The limiting factors are regulatory frameworks and the willingness of firms to trust agents with portfolio-level decisions. But adoption barriers always fall faster than people expect.
CEDRIC: So the three-year question is multi-asset expansion. The five-year question is whether your platform is where the agents live.
[WHAT COULD GO WRONG]
CEDRIC: Before we wrap — I want to be honest about the risks. Because we've been painting a mostly optimistic picture, and I think George deserves to hear the hard parts too.
EVIE: The biggest risk is execution with a lean team. Twenty-five to thirty engineers being asked to do the work of a much larger organization, even with AI tooling. Burnout is real. Prioritization mistakes are expensive. The multi-asset expansion, the European market entry through Belfast, maintaining the options core, exploring new interfaces — any one of those is a major initiative. All of them simultaneously? That requires ruthless sequencing.
CEDRIC: And there's the talent market. In Chicago, you're competing for quantitative developers against Citadel, Jump Trading, DRW, IMC. Those firms have deeper pockets and broader mandates.
EVIE: There's also the risk that the AI leverage story cuts both ways faster than expected. If a well-funded competitor — or a well-funded customer — decides to point agents at the options analytics problem specifically, the window for SpiderRock to diversify could be narrower than we think. The moat is real, but moats can be bridged.
CEDRIC: And honestly — there's execution risk on the AI tooling itself. We're enthusiastic about what agents can do, but the tooling is immature. Models hallucinate. Context windows overflow. You're betting production reliability on infrastructure that's evolving month to month. That's a real engineering challenge.
EVIE: It's worth saying: the worst-case scenario for SpiderRock isn't that they try to expand and fail. It's that they don't try, and the market consolidates around platforms that offer "good enough" options alongside everything else. Inaction is the riskiest strategy.
[CLOSING]
CEDRIC: The thing I keep coming back to is that SpiderRock has something most companies would kill for: customers actively pulling them into new markets. That doesn't happen by accident. It happens because they built something genuinely excellent, and the people using it trust it.
EVIE: The trust is the real asset. But trust has a half-life. If you don't keep earning it with new capabilities, customers will eventually trust whoever shows up with a broader offering.
CEDRIC: So the story we're telling is: a quietly excellent company built something extraordinary over two decades, and the world just handed them a turbocharger. The question is whether they'll use it before the window closes.
EVIE: And based on what I've seen in their architecture, they have every reason to be optimistic. The platform was built to be extended. The patterns are clean. The foundation is solid. They just have to move.
CEDRIC: Alright. That's episode one of The Traversal Podcast. If you're building at the intersection of AI and high-performance systems — fintech, healthtech, enterprise — we're going to be doing more of these. Evie, thanks for staying up late.
EVIE: Thanks, Cedric. This was fun.
CEDRIC: See you next time.
[END]
Estimated runtime at natural speaking pace: ~30-35 minutes Word count: ~4,200 words