The Pivot to AI: Building GrackerAI

After building LoginRadius into a platform serving more than 1 billion users, I did something that surprised a lot of people, including myself. In 2024 I started a new company in a different space. GrackerAI is not an identity management platform. It does Generative Engine Optimization (GEO) for B2B cybersecurity vendors: helping them earn citations and recommendations from AI engines.

This chapter is about how that pivot happened, what it takes to apply domain expertise to a new market, and the realities of serial entrepreneurship that nobody prepares you for.

---

Recognizing the GEO Opportunity

The insight came from an unexpected place: content marketing. LoginRadius had invested heavily in it for years, with a large library of guides and tutorials, and I had been publishing on identity and security for just as long.

Then the referral data started to change. Visits began arriving from AI assistants such as ChatGPT and Perplexity, a source that had not existed a year or two earlier.

I started testing. I asked ChatGPT questions about CIAM platforms. Sometimes a given vendor was cited. Sometimes it was not. The inconsistency bothered me. Strong, detailed content did not guarantee a citation, and AI citation looked random.

It was not random. It was based on patterns I did not yet understand: content structure, authority signals, entity clarity, and freshness markers. The same patterns I would later codify into the GEO framework.

The Market Timing Signal

Several signals converged that told me this was a company-sized opportunity, not just a marketing tactic:

Market Timing Signals for GEO
================================

1. ChatGPT reached 200M weekly users
   (August 2024)
   -> Massive adoption, not a niche tool

2. Google rolled out AI Overviews to
   all US users (May 2024)
   -> The dominant search engine was
      changing its core product

3. Enterprises were already on board
   -> OpenAI said 92% of the Fortune 500
      used its products

4. Few established tools for GEO
   -> SEO tools had existed for 20 years
      but almost nothing served GEO

5. Content marketers were confused
   -> Demand for answers was visible
      in every marketing community

6. The pattern matched CIAM in 2013
   -> Category did not have a name yet
      but the problem was growing fast

Sources for the signals: OpenAI's 200 million weekly users and Fortune 500 figure were reported by VentureBeat in August 2024. Google's US rollout of AI Overviews was covered by 9to5Google in May 2024. The adoption curve has only steepened since: Sam Altman said ChatGPT had 800 million weekly active users in October 2025.

When I applied the Seven Signals Framework from Chapter 2 to the GEO opportunity, it scored 6 out of 7, about where CIAM stood when I started LoginRadius. I recognized the pattern.

Building an AI-Native Product

Building GrackerAI was fundamentally different from building LoginRadius. LoginRadius was an infrastructure product: APIs, SDKs, and backend services. GrackerAI is an AI-native product. It uses large language models, semantic analysis, and real-time data processing as core capabilities, not just features.

What "AI-Native" Actually Means

Dimension	Traditional SaaS (LoginRadius)	AI-Native SaaS (GrackerAI)
Core logic	Deterministic code	LLM-powered analysis and generation
Data processing	Structured data in databases	Unstructured text processed semantically
Output consistency	Same input = same output	Same input may produce different outputs
Testing approach	Unit tests, integration tests	Evaluation sets, human review, statistical testing
Infrastructure costs	Compute scales linearly	LLM inference costs dominate and are non-linear
Development speed	Predictable sprint velocity	Variable: some features take hours, others weeks
User expectations	Exact, reliable behavior	Tolerance for approximate, intelligent behavior

The AI Product Development Challenges

Non-determinism. Traditional software does the same thing every time. AI products can produce different outputs for the same input. This makes testing, quality assurance, and customer support fundamentally harder. We had to build evaluation frameworks that tested output quality statistically rather than deterministically.

Cost management. LLM inference is expensive. Every API call to a large language model costs money. Building a product where the core functionality requires LLM calls means your cost of goods sold is directly tied to usage volume. We had to architect aggressively for cost efficiency: caching results, using smaller models where possible, and batching requests.

Prompt engineering as product development. In traditional software, you write code that executes logic. In AI-native products, you write prompts that guide model behavior. Prompt engineering became one of our core product development disciplines: iterating on prompts the way traditional teams iterate on code.

Rapid model evolution. The LLM landscape changes monthly. A capability that required complex engineering in Q1 might be available as a single API call by Q3. We learned to hold our architecture loosely and rebuild frequently, using abstractions that let us swap models without rewriting the product.

The Future of Identity: Beyond Traditional CIAM

Building LoginRadius taught me the foundational importance of customer identity and access management. LoginRadius solved a critical B2B authentication problem: giving legacy and enterprise companies a reliable way to manage customer identities at scale. But having spent years at the intersection of identity, security, and now AI, I see clearly that the future of identity demands far more than traditional CIAM ever offered.

The next generation of identity systems will not just authenticate users. They will be intelligent, adaptive, and capable of operating in a world where the boundaries between human and machine identity are increasingly blurred.

AI-Powered Identity Management

AI is fundamentally reshaping how identity systems work. Traditional CIAM platforms rely on static rules. If a login comes from a new device, trigger MFA. If a user fails three password attempts, lock the account. These rule-based systems are predictable, which makes them both reliable and exploitable.

AI-powered identity systems operate differently:

Capability	Traditional CIAM	AI-Powered Identity
Threat detection	Rule-based triggers	Behavioral analysis detecting anomalies in real time
Authentication decisions	Binary (pass/fail)	Risk-scored, contextual, continuous
Fraud prevention	Pattern matching against known threats	Predictive models identifying novel attack vectors
User experience	One-size-fits-all security flows	Adaptive flows that adjust friction based on risk
Identity verification	Document checks at onboarding	Continuous verification through behavioral biometrics

Behavioral biometrics (the way you type, move your mouse, hold your phone, and navigate an application) create a continuous identity signal. In my view it is much harder to spoof than a password, and AI makes it practical at scale.

Post-Quantum Cryptography and Identity

Every identity system built today relies on cryptographic primitives that quantum computers will eventually break. RSA, ECC, and the Diffie-Hellman key exchange that underpin TLS, JWT tokens, and certificate-based authentication are all vulnerable to Shor's algorithm running on a sufficiently powerful quantum computer.

This is not a theoretical concern for the distant future. Adversaries can capture encrypted data today and decrypt it once quantum capabilities mature, the "harvest now, decrypt later" threat that NIST cites as a reason to migrate now. Identity systems that issue long-lived tokens, certificates, or credentials need to begin their migration to post-quantum cryptographic algorithms now.

In August 2024 NIST finalized its first three post-quantum standards: FIPS 203 (ML-KEM, derived from CRYSTALS-Kyber) for key encapsulation, and FIPS 204 (ML-DSA, from CRYSTALS-Dilithium) and FIPS 205 (SLH-DSA, from SPHINCS+) for digital signatures. Forward-thinking identity platforms are building crypto-agility: the ability to swap cryptographic algorithms without rebuilding the entire system. Any identity infrastructure built today without crypto-agility is accumulating technical debt that will become a security liability.

Passwordless Authentication as the New Standard

The password is finally dying. Not because security professionals have said it should for decades, but because viable alternatives have reached mainstream adoption. FIDO2, WebAuthn, and passkeys represent a fundamental shift in how authentication works.

Passkeys, backed by Apple, Google, and Microsoft through the FIDO Alliance, are making passwordless authentication available to billions of users. They are phishing-resistant by design, require no memorization, and provide a better user experience than passwords. The question is no longer whether passwordless will become standard but how quickly organizations will migrate.

For identity platforms, this shift is existential. Systems designed around password storage, hashing, and reset flows need to fundamentally rearchitect for a world where the primary credential is a cryptographic key pair managed by the user's device. LoginRadius and platforms of its generation were built for the password era. The next generation of identity systems will be built for a passwordless world.

Machine Identity: The Fastest-Growing Identity Challenge

As AI agents, IoT devices, microservices, and automated workflows proliferate, machine-to-machine identity is becoming a larger challenge than human identity management. Palo Alto Networks' 2026 Identity Security Landscape puts the ratio at 109 machine identities for every human, up from 82 in CyberArk's 2025 edition, which credited AI adoption for much of the growth.

AI agents that browse the web, make API calls, execute transactions, and interact with other AI agents need their own identity layer. How do you authenticate an AI agent? How do you authorize its actions? How do you audit what it did and why? How do you revoke access when an agent is compromised or decommissioned?

The answers are maturing fast. Microsoft made Entra Agent ID generally available in May 2026, registering agents as directory identities. OAuth token exchange and the MCP authorization spec give agents delegated, scoped access. The Agent Security Map tracks the vendors building this layer. The traditional CIAM model, designed for humans logging into web applications, still does not cover it on its own. Machine identity requires:

• Non-interactive authentication that does not rely on human presence
• Granular, dynamic authorization that can adjust in real time based on context
• Attestation and provenance so you can verify not just who is calling but what code is running and where
• Short-lived credentials that minimize the blast radius of compromise
• Identity federation across AI ecosystems where agents from different organizations need to interact securely

Intelligent Identity: Systems That Learn, Adapt, and Predict

The most significant shift is from identity management to intelligent identity orchestration. Traditional CIAM manages identities: it stores them, authenticates them, and enforces policies. Intelligent identity systems do more: they learn from every interaction, adapt their behavior in real time, and predict threats before they materialize.

Imagine an identity system that:

• Recognizes that a particular user's behavior pattern has shifted and proactively adjusts security requirements before any rule is triggered
• Predicts which accounts are likely to be targeted in an upcoming credential-stuffing attack based on dark web intelligence and behavioral signals
• Automatically adjusts authentication flows across an entire organization based on the current threat landscape
• Understands the relationships between identities, human and machine, and detects lateral movement attempts by analyzing identity graph anomalies

This is the direction identity is heading. LoginRadius solved the foundational problem of B2B customer identity management, and that foundation remains critical. But the future belongs to identity systems that are not just secure but intelligent: systems that treat identity as a dynamic, context-aware, continuously evaluated signal rather than a static credential check.

For founders building in the identity space today, the opportunity is not in building another CIAM platform. It is in building the intelligent identity layer that connects authentication, authorization, threat detection, machine identity, and post-quantum security into a unified, AI-powered system.

Applying Domain Expertise to a New Market

The transition from CIAM to GEO was not as discontinuous as it appears. Several skills and knowledge areas transferred directly:

What Transferred

Domain Expertise from CIAM	Application to GEO
Enterprise sales process	Same buyers (B2B SaaS marketers work with the same companies)
Content marketing execution	Content is both our product's domain and our marketing channel
B2B SaaS business model	Subscription pricing, expansion revenue, customer success
Technical credibility	Security and engineering background builds trust
Understanding of AI/ML	Identity systems use ML for fraud detection and risk scoring
Category creation experience	GEO is another category that needs to be defined and established

What Did Not Transfer

CIAM Expertise	Why It Did Not Apply to GEO
Security compliance (SOC 2, ISO 27001)	GEO does not handle sensitive data the same way
Authentication architecture	Completely different technical domain
Identity federation protocols	No relevance to content optimization
Enterprise security sales cycle	GEO sales cycles are shorter, less scrutiny-heavy
Multi-tenant security architecture	Different infrastructure requirements

Serial Entrepreneurship Lessons

Building GrackerAI after LoginRadius has taught me lessons about serial entrepreneurship that I could not have learned from a single company.

Lesson 1: Speed Comes From Pattern Recognition, Not Shortcuts

GrackerAI moved from idea to initial product much faster than LoginRadius did. Not because we cut corners, but because we recognized patterns earlier. The difference is not talent or resources. It is knowing what to build first, what to skip, and what to validate before building. Modern AI tooling helped too: a first prototype now takes weeks, not months.

Lesson 2: Your Network Compounds

The relationships I built over a decade in security and B2B SaaS were GrackerAI's biggest unfair advantage. They shaped who we could reach for early feedback, who we could ask for advice, and who already read what I published.

Your network does not just help. It accelerates everything. Every year you spend building genuine relationships in your industry makes your next venture easier.

Lesson 3: The Market Does Not Care About Your Track Record

Having built a successful company helps with fundraising, hiring, and networking. It does not help with product-market fit. The market does not give you credit for previous success. GrackerAI's early customers did not buy because I built LoginRadius. They bought because GrackerAI solved their problem.

This is a humbling and important lesson. Past success can create dangerous overconfidence. Every new product still needs to prove itself on its own merits.

Lesson 4: Build Different, Not Bigger

The temptation for serial founders is to build the next company bigger and faster than the last one. Raise more money. Hire more people. Launch with more features.

At GrackerAI, I did the opposite. Smaller team. Lower burn rate. Narrower initial scope. Faster iteration cycles. The goal was not to replicate LoginRadius at twice the speed. It was to build a different kind of company that played to the strengths of the current market.

Lesson 5: What LogicBalls Taught Me

Alongside Start with Identity, I run LogicBalls, a practitioner community for people learning to get real value from AI. It is not a venture-scale company, and it was never meant to be. Watching how people outside engineering use AI there taught me where they trust the output and where they do not, and that informed GrackerAI's product design.

Not every venture needs to be a billion-dollar company. Sometimes the most valuable thing you build is the knowledge that feeds your next venture.

The AI Entrepreneur's Advantage

Building a company in the AI era offers advantages that did not exist when I started LoginRadius:

Faster prototyping. What took months of engineering can now be prototyped in days using LLM APIs and AI-powered development tools.

Lower infrastructure costs. Cloud-native architecture and managed services mean you can launch a production product with a fraction of the infrastructure investment required a decade ago.

Accessible AI capabilities. You do not need a PhD in machine learning to build AI-powered products. Pre-trained models, APIs, and frameworks make sophisticated AI accessible to application developers.

Larger addressable markets. AI is creating new categories and expanding existing ones. The market for GEO tools barely existed before 2024. Now GEO Compass profiles a whole field of AI visibility vendors.

What I Would Do Differently

If I were starting an AI company from scratch today, I would do three things from the start:

1. Hire a designer on day one. AI products require exceptional UX because users need to trust outputs they cannot fully verify. A designer-founder partnership from day one shortens the path to product-market fit.

2. Build the community before the product. I would start a community (newsletter, Discord, events) months before launching the product. Build the audience first, then build the product for that audience.

3. Charge from day one. A generous free tier is tempting. Paying customers give better feedback, have higher engagement, and create real validation. Free users create the illusion of traction without the reality of product-market fit.

The journey from LoginRadius to GrackerAI reinforced my belief that the best founders are the ones who cannot stop building. Not because they enjoy the startup grind (nobody does), but because they keep seeing problems that need solving and cannot resist the pull of building the solution.

The final chapter distills my years of building into the advice I wish someone had given me when I started.