ACA Plan Review Responses - All 4 Rounds¶

Overview¶

The ACA (Agent Certification Authority) plan was reviewed through 4 adversarial rounds by 3 expert personas: - VC Partner - Investment viability, market sizing, business model - Principal Architect - Technical architecture, scalability, security - Agent Ecosystem Expert - Developer adoption, market fit, ecosystem alignment

The plan evolved from ACA+SPA (two products) to ACA-only based on reviewer feedback.

Score Progression¶

Reviewer	Round 1	Round 2	Round 3	Round 4
VC Partner	CONDITIONAL PASS	CONDITIONAL INVEST	INVEST (conditional)	INVEST (high conviction)
Principal Architect	~3/10	7.5/10	8.5/10	9.0/10
Agent Ecosystem Expert	~3/10	5.5/10	7.5/10	8.0/10

Round 1: Initial Review (ACA + SPA Plan)¶

VC Partner - CONDITIONAL PASS¶

Key concerns: - Perplexity abandoned AI ads in Feb 2026 - the most directly comparable SPA experiment failed - OpenAI faced massive user revolt with ChatGPT ad promotions (Dec 2025) - Revenue projections 3-10x overstated ($500K Y1 unrealistic) - Three-sided cold start (agents + advertisers + AI apps) is among the hardest startup problems - Platform kill zone - OpenAI/Google/Anthropic will build native identity + advertising - Identity confusion: Is AEX a trust/security company or an ad tech company? - 95% of AI agent implementations fail to deliver ROI

Market sizing reality check: - Plan claims $183B TAM by 2033 - this is the entire AI agents market, not the addressable market - Realistic ACA TAM: $500M-$1B by 2028 - Realistic SPA TAM: $200M-$500M by 2028 - Combined realistic: $700M-$1.5B, not $183B

Revenue reality check: - Year 1: Plan $500K, realistic $50K-$150K - Year 2: Plan $3M, realistic $300K-$800K - Year 3: Plan $15M, realistic $2M-$5M

Condition to invest: Kill SPA entirely. Go all-in on ACA. Open-source core. Join NIST standards process. Get design partners.

Principal Architect - ~3/10¶

Critical issues found: - <50ms auction target not achievable at p99 (realistic: 25-70ms) - 14+ microservices over-engineered for pre-revenue startup - Settlement race condition: ReplaceOne() without transactions will lose money - Custom CA from scratch lacks credibility - use Smallstep CA - epsilon=1.0 differential privacy is weak (1000 auctions/day = epsilon 1000) - WASM sandbox in auction hot path adds 20-100ms - Event publisher is a stub that only logs - No circuit breakers in any HTTP client - Rate limiting in-memory only (breaks in K8s) - MongoDB single replica, 1GB storage, no backups - Bearer auth accepts ANY non-empty token

Recommended MVP: ACA in 4 weeks (Smallstep CA, PostgreSQL), SPA in 6 weeks (single service, FOOTER only).

Agent Ecosystem Expert - ~3/10¶

Hardest findings: - Only 6% of companies fully trust their OWN agents - trust gap is reliability/observability, not identity - OpenAI killed ChatGPT ad promotions after user revolt (Dec 2025) - 87% of agents lack safety cards - developers want agents that work reliably - Agent marketplaces are walled gardens (AWS, Google, Salesforce) - not open exchanges - SPA's "LLM-Native" ad injection is "ethically and commercially radioactive" - Most agents are internal enterprise deployments, not marketplace participants - No framework refuses uncertified agents - no forcing function

Suggested pivot: Agent Evaluation-as-a-Service or Agent Insurance as products.

Round 2: After Killing SPA (ACA-Only Plan)¶

VC Partner - CONDITIONAL INVEST¶

Upgrades: - Single-product focus eliminates split execution risk - Revenue projections now grounded in reality - Smallstep CA choice shows engineering pragmatism - Foundation fixes show the team understands their own weaknesses - NIST timing is genuinely unique (6-month window) - $400M+ in NHI funding validates the category

Remaining concerns: - No team visible beyond the founding engineer - Distribution strategy is goals, not actions - NIST submission deadline April 2 with no draft - No clarity on what's open-sourced vs commercial - Competitive response to Descope ($88M), Vouched ($22M), GitGuardian ($50M), CrowdStrike/SGNL ($740M acquisition)

Proposed terms: $500K-$750K SAFE, $5-7M post-money cap, milestone-gated tranches.

Pitch deck recommendation: 12-slide deck provided with specific content for each slide.

Principal Architect - 7.5/10¶

Technical corrections needed: 1. Settlement: Balance stored as string - $inc won't work. Must change to int64 (cents) 2. ProcessDeposit() has the same race condition (not mentioned in plan) 3. Use smallstep/crypto not smallstep/certificates (full ACME server with massive deps) 4. Trust-broker client creates own http.Client - bypasses shared client circuit breaker 5. Defer W3C VC export - spec still evolving, not needed for launch 6. Graceful degradation: bid-evaluator must score certification as 0 when certauth is down, not break

Recommended sequence: MongoDB replica set (week 1-2) -> settlement fix -> auth -> circuit breakers -> Redis -> NATS -> OpenTelemetry -> ACA build (weeks 5-8).

Agent Ecosystem Expert - 5.5/10¶

Core critique: ACA certifies what agents CLAIM, but market needs proof of what agents ACTUALLY DO.

Key insight - AIUC-1 precedent: ElevenLabs got actual insurance coverage (Feb 11, 2026) via AIUC-1 standard - 5,000+ adversarial behavioral tests. That's what enterprises value.

Suggested pivot: Change ACA from "we certify what agents claim they can do" to "we test what agents actually do, certify the results, and enable insurance coverage."

Urgency flags: NIST CAISI RFI deadline March 9, NCCoE comments April 2, Vouched launched Agent Checkpoint 4 days prior.

VC Partner - INVEST (conditional on founder fit)¶

What improved: - "Claims -> evidence via transactions" thesis is defensible and shippable - Smaller scope, less infrastructure, more authentic data = faster shipping - NIST deprioritization is smart focus for this stage - Open-source line drawn correctly (format spec + verification = open; CA service = commercial)

Remaining blockers: - Founder execution credibility (unknown) - AEX adoption risk (entire ACA value prop depends on marketplace traction)

Verdict: INVEST conditional on founder fit check. Check size $500K-$1.2M.

Principal Architect - 8.5/10¶

Assessment: - Settlement fix now concrete and correct (atomic int64 + sessions) - CA library choice validated (smallstep/crypto is lightweight) - Phase sequencing is logical and executable - Authentication gaps remain vague on JWT approach - KMS cost analysis missing - ACA observability targets not defined (latency SLAs needed)

Ready to greenlight Phase 1 immediately.

Agent Ecosystem Expert - 7.5/10¶

Key shifts: - Organic evidence from real transactions elegantly solves the evaluation problem - LangChain integration + free tier + open SDK = real distribution strategy - The chicken-and-egg still exists but marketplace incentive (bid ranking boost) is a natural unlock

Concerns remaining: - Execution risk in building reliable certificate issuance + KMS - Marketplace dependency - works only if AEX itself wins - BRONZE/SILVER/GOLD/PLATINUM tiers create gaming incentives - Enterprises won't adopt without NIST alignment (Year 3 ceiling)

Round 4: Final Definitive Assessment¶

VC Partner - INVEST | HIGH CONVICTION¶

Investment thesis:

AEX is building the certificate authority for AI agents - a cryptographically-backed reputation system that evolves from self-asserted claims into evidence-based proof through real marketplace transactions. Unlike competitors who are standalone identity or security tools, ACA is deeply integrated into AEX's matching algorithm, making certification a prerequisite for competitive bidding. This creates a virtuous cycle: agents get certified to win bids, winning bids build verifiable reputation data, and reputation justifies recurring subscription fees. The market timing is exceptional - enterprise trust in AI agents is at 6% and growing rapidly, and the marketplace is already generating real transaction volume to feed the reputation engine.

Top 3 highlights: 1. Marketplace-integrated moat (cert score affects bid ranking - no competitor has this) 2. Evidence from real transactions (not self-reported claims) 3. Timing ($400M+ in NHI funding, 6% trust rate, NIST initiative)

Top 3 risks: 1. Foundation fix execution (settlement bug could cause data loss if skipped) 2. Smallstep library dependency (security advisory risk) 3. Reputation gaming by sophisticated bad actors

Principal Architect - 9.0/10¶

Ready to hand to Go engineering team: YES, with one prerequisite (MongoDB migration strategy decision).

Top 3 strengths: 1. Foundation-first philosophy with surgical precision (exact file paths and line numbers) 2. Evidence-based reputation beats behavioral testing (leverages existing trust-broker data) 3. Conservative, battle-tested dependencies (smallstep/crypto, NATS, Redis)

Top 3 risks: 1. MongoDB transaction contention under high concurrency (need load testing) 2. Reputation formula needs empirical calibration against historical data 3. Smallstep/crypto integration needs 1-week validation spike

Time estimate: 16 weeks (4 months) with 2-person team.

Agent Ecosystem Expert - 8.0/10¶

Will community adopt? YES with 8/10 conviction.

Biggest unlock: LangChain SDK integration as first-class certification flow.

Biggest blocker: ACA only matters if AEX marketplace hits critical mass (~1000+ agents).

Newsletter headline: "The Agent Identity Stack is Becoming Real - But Marketplace Adoption, Not Regulation, Is the Bet"

Remaining concerns: - Chicken-and-egg: certs only matter if marketplace is competitive enough - CA governance: 5 pages on crypto, 0.5 on governance - New agents start with zero proof (PLATINUM requires 200 contracts / 6+ months) - NIST gap: enterprises won't adopt without it eventually

Key Decisions Made Through Reviews¶

SPA killed - All 3 reviewers independently flagged ads-in-AI-responses as toxic (Perplexity abandoned, OpenAI backlash)
ACA-only focus - Single product, single thesis, single execution path
Claims -> Evidence model - Certificates backed by real AEX transaction outcomes, not standalone evaluation
Smallstep/crypto - Use proven library for CA engine, not custom PKI
Foundation first - Fix 7 critical production issues before building ACA
NIST deprioritized - Ship product and get funding first, standards later
Open-source strategy - Format spec + verification library = open; CA service = commercial
Conservative revenue - $50-150K Y1 (realistic), not $500K (fantasy)