Skip to content
Graph of Plan

Agent-native execution is the platform shift

AI agents are getting deployed into production. Most of them run in environments designed for humans—with human-sized permissions, human-oriented interfaces, and audit trails that assume a person is at the keyboard.

This seems like a problem worth solving.

The bet

We think purpose-built infrastructure for AI agents will be more useful than retrofitting human tools. Specifically:

  1. Sandboxed execution: Agents need environments they can’t accidentally escape from
  2. Audit trails: When an agent does something, you need to know what and why
  3. Typed interfaces: Agents work better with structured APIs than raw HTTP
  4. Capability-based permissions: Fine-grained, revocable, auditable

This might be wrong. The existing tools might be good enough. But we’re building as if they’re not.

Why now (and why it might be too early)

Arguments for:

  • Language models can write useful code (not perfect, but useful)
  • Context windows fit entire codebases
  • Tool use works reliably enough for production
  • Costs are dropping fast

Arguments against:

  • Most teams aren’t deploying autonomous agents yet
  • The compliance/audit market is nascent
  • “Agent-native” might be a distinction without a difference

We’re betting the “for” column wins, but we’ve hedged by building products that work even if the platform thesis is wrong.

Market Opportunity

We’re not chasing a giant TAM. We’re building for a specific niche: teams deploying AI agents who need audit trails and sandboxed execution.

The Niche (Bottoms-Up)

Nomos Cloud: AI teams needing compliance

UK/US companies deploying AI agents in production:     ~5,000
× Need audit trails for compliance/governance:         × 40%
× Can pay £150+/mo for tooling:                       × 50%
= Addressable customers:                               1,000


At £150 ARPU + 10% enterprise at £25K ACV:
→ Niche opportunity: ~£4M ARR

Murphy: Agencies with delivery pain

UK/US digital agencies (5-50 staff):                   ~15,000
× Have multi-project delivery complexity:              × 30%
× Actively seeking delivery tooling:                   × 20%
= Addressable customers:                               900


At £600 blended ARPU:
→ Niche opportunity: ~£6M ARR

SmartBoxes: Non-developers wanting AI automation

UK/US solopreneurs/SMBs who would pay for AI tools:    ~500,000
× Specifically want "AI that builds for me":           × 2%
× Will pay £30+/mo:                                    × 30%
= Addressable customers:                               3,000


At £37 ARPU:
→ Niche opportunity: ~£1.3M ARR

P4gent: Individuals managing supplier spend

UK/US self-employed managing 10+ suppliers:            ~2M
× Aware of subscription/supplier management problem:   × 5%
× Will pay for tooling:                               × 10%
= Addressable customers:                               10,000


At £22 blended ARPU:
→ Niche opportunity: ~£2.6M ARR

Total Niche: ~£14M ARR

This isn’t a massive market. That’s the point. It’s a niche we can actually win with a small team, where:

  • Customers have acute pain (compliance risk, delivery slips, time poverty)
  • Willingness to pay is proven (competitors exist at similar price points)
  • We have differentiation (Cloudflare-native, agent-first architecture)

Revenue Targets

TimeframeMRR TargetARRNiche Penetration
Year 1£100K£1.2M9%
Year 2£400K£4.8M34%
Year 3£800K£9.6M69%

Year 1 breakdown by product (M12 MRR):

  • Nomos Cloud: £62K (61%) — highest ARPU, enterprise deals
  • Murphy: £19K (19%) — agency wedge
  • SmartBoxes: £17K (17%) — platform foundation
  • P4gent: £3K (3%) — consumer experiment

Caveats on These Numbers

These estimates involve guesswork. The conversion percentages are made up. The ARPU assumptions might be wrong. The total addressable customers are rough approximations.

What we’re reasonably confident about:

  • The niche is small (definitely not a billion-dollar TAM)
  • Customers in these segments do pay for tooling (competitors exist)
  • The problems are real (we’ve experienced them)

What we’re uncertain about:

  • Whether our products solve the problems better than alternatives
  • Whether the market timing is right
  • Whether we can execute well enough to capture meaningful share

Why This Niche, Not Something Bigger?

  1. We can actually win it: A £14M niche with 4 focused products is achievable for a small team
  2. If we’re wrong, we find out quickly: Small markets give fast feedback
  3. Competitors are weak here: Big players don’t care about £14M markets
  4. It might grow: If agent deployment accelerates, the niche grows with it (but this is uncertain)

Justified By