Market landscape
Reliability & oversight for the agentic economy
The vertical: catching what agent-driven integrations break, and giving a human the yes/no on what agents do. Sized, growing, and structurally unsolved at the decision surface.
Size & growth
$6.8B
LLM / agent observability by 2029
$1.97B (2025) · 36.5% CAGR
$50.3B
AI agent market by 2030
$7.63B (2025) · 45.8% CAGR
$11.6B
"Securing AI" budget by 2029
$2.8B (2026) — the underfunded side
$234B
enterprise spend exposed to agentic arbitrage
Gartner · by 2030
Key players, by layer
Enforcement
custodian-kernel (OSS, SMS-only) · Numbat (Perplexity) · dcg (deterministic guard) · Stripe Link (agent wallet, single-rail)
Observability
Langfuse (OSS) · LangSmith (LangChain) · Arize · Braintrust · Helicone · Datadog
Gateways
Portkey — routes and guardrails LLM calls (Nigraan consumes, doesn't compete)
Webhook reliability
Hookdeck · Svix · ngrok — the plumbing underneath (SupaHooks plays here)
Trust / governance
Vanta · SafeBase · Waxell · Omnara · HumanLayer — trust centers & approvals, each scoped to one surface
Why now
- Adoption outpaces control. 81% of teams deploy agents into production; only ~14% have full security approval; 48% run with zero oversight; 88% report an agent incident in the past year.
- Regulation just bit. The EU AI Act's human-oversight mandate (Art. 14) became enforceable in August 2026 — meaningful human control is now a legal requirement.
- The failure modes are real and current. A misaligned agent published a hit piece on a maintainer to reverse a rejection; a "trusted" agent can cross security boundaries just by trying too hard to finish its task.
- Per-seat is collapsing. Agents wipe out the human users of SaaS and become the users themselves — value shifts to governance, data, and outcomes.
- Money is moving to governance. Security, governance, and compliance is the "boring but massive" destination — $48.5B goes to AI-amplified security vs $2.8B to secure the agents themselves.
Where Nigraan sits
Three teams solved enforcement. Nobody solved the human decision.
Every rail — kernels, gateways, observability, payments — routes approvals to Slack, SMS, or a web dashboard. There is no cross-vendor, mobile, calibrated surface where a person decides. Nigraan is that layer: reliable plumbing (SupaHooks) + a standard runtime (agent-harness) + a mobile human yes/no that stays high-signal instead of becoming a rubber stamp.