Founder with exits
Leadership roles at Vault Rooms, Copilot Communications, Doctors Access and Vital Care Services.
01Pre-seed / Investor brief
From objective
to production.
Under your control.
Legatus coordinates AI coding agents from objective to production, enforces requirements, and checks their work before it advances.
Gary SomerhalderFounder & CEO / 16+ years in enterprise · Leadership at 4 acquired companies
Objective → production
2 years building
Built over two years with our own capital, before this pre-seed raise.
Real client projects
Tested inside Centurion Software Group on our own client projects.
Paid pilot: Swordsweeper · Raising $1.5M pre-seed
Development history and internal testing are founder-reported. Prior-company exits are founder experience; pilot outcomes remain to be measured.
02The initial customer
More code. The same accountability.
AI agents can produce the change. Engineering leaders still have to establish that it meets the requirements and is ready to advance.
VP Engineering / CTO
B2B software teams with 20–200 engineers, an active coding-agent rollout, and meaningful release requirements.
Reconstructing “done”
Reconcile requirements, test results, reviews, and exceptions across agent sessions before accepting the work.
Scaling beyond one agent
A broader rollout needs a consistent acceptance process. The pilot measures the verification work that process can remove.
First customer discovery question: how much work happens between an agent saying “done” and a team accepting the change?
Gary SomerhalderFounder & CEO / Legatus AI
03Why this founder
Four acquired companies. Enterprise depth.
Gary Somerhalder brings CEO, CTO and IT leadership experience across healthcare, communications and secure software to Legatus.
Vault Rooms
Chief Technology Officer
Jan 2017 – Jan 2018Copilot Communications
Chief Executive Officer
Feb 2014 – Jan 2016Doctors Access
Chief Technology Officer
Jan 2013 – Jan 2014Vital Care Services
Director of Information Technology
Oct 2009 – Oct 201116+ years
Experience across healthcare, finance, travel, and secure data workflows—including EHR and virtual data room platforms.
320% ROIC
CTO and turnaround leadership after a bankruptcy IP acquisition; maintained service for 150+ medical practices. Prior-company exit outcome.
2 years building
Self-funded development, tested inside Centurion Software Group on its own client projects. Leading the runtime and verification architecture.
Roles and acquisition statuses are founder-reported. Dates shown are role tenure, not acquisition dates. The 320% ROIC relates to Doctors Access, not Legatus.
04Founder experience / before Legatus
Enterprise experience. From strategy to execution.
Through Centurion Software Group and GMS Ventures, Gary has built teams, platforms and partnerships; led restructurings and acquisitions; and connected technology with business operations and capital.
An operator’s perspective
Strategy, growth, capital, teams and operating models—from venture-backed startups to global enterprises.
Architecture through delivery
Senior software engineering, enterprise platforms and agentic AI expertise connect the vision to working systems.
Systems that face scrutiny
Healthcare, finance, communications, travel and technology inform the focus on requirements, accountability and reliable delivery.
Representative organizations across prior work and engagements
Representative organizations from Gary’s published background; relationships vary by engagement. Founder experience before Legatus, not a Legatus customer list.
05The market shift
AI adoption is ahead of trust.
Stack Overflow’s 2025 Developer Survey shows widespread AI-tool adoption alongside a substantial verification problem.
84%
Use or plan to use AI tools in development. This includes AI tools broadly, not only autonomous coding agents.
46%
Distrust AI-tool accuracy, compared with 33% who trust it. Adoption does not remove the need to check the output.
Make acceptance repeatable
Give teams a consistent way to connect requirements, checks, and policy to the decision that work can advance.
Source: Stack Overflow Developer Survey 2025; question-specific respondents. These figures establish context, not buyer counts or demand for Legatus.
06Implementation evidence
A working runtime. Built to enforce delivery.
A nine-crate Rust runtime implements workflow gates, evidence evaluation, signed proof chains and durable checkpoints. Customer validation now tests these mechanisms in real delivery workflows.
IMPLEMENTED IN REPOSITORY
Evidence before progression
Phase completion requires a durable checkpoint and evidence evaluation. Unknown workflow and phase submissions are rejected.
Pilot measure: bypass attempts and false blocks on supported execution paths.
Inspect source ↗Repository access may be requiredImplementation inspected in the repository. Source presence establishes the build; pilot evidence must establish reliability and production coverage.
07The control plane / Interactive illustration
Delegate the work. Keep command.
Set direction. Coordinate agents. Enforce policy. Verify the work. Explore the control plane from objective to production.
The agent says done. Who checks?
Same task. Same failure. Start with an agent session that has no Legatus gate.
AT STAKE / ILLUSTRATIVE SCENARIO
14 scripted scenarios · no live execution
Same scenario. Choose a side to run it.
Illustrative labor estimate · one incident · USD
Assumes existing checks miss the issue versus Legatus catching it before release or handoff.
See cost assumptions
Planning inputs, not measured results. Existing review may also catch the issue. Labor only; software, compute and business losses are excluded.
● ● ●~/centurion/account-serviceClaude Code
✳
Claude CodeCoding session / interactive demo
NO LEGATUS GATE completion is an agent claim
▋Ready · mandate defined
WHAT THIS COULD MEAN FOR ONE TEAM Illustrative assumptions · not measured savings
66.7hreview capacity recovered / month
$8,000monthly capacity value
$5,700net modeled value after added costs
See the math · change assumptions 3.5× modeled value / cost
200 tasks × (30 − 10) review minutes ÷ 60 = 66.7 hours / month.
Capacity value is not cash savings. Include human review, exceptions, and false holds in the “with” time. $2,000/month reflects proposed pricing, not pilot terms. Existing agent spend cancels out only if unchanged; enter extra checks and repair costs above. No avoided incidents or revenue uplift is assumed.
What the pilot must prove
Human minutes / accepted taskMatched baseline vs. Legatus, including intervention and rework.
Cost / accepted taskHuman time + agent execution + verification + allocated subscription.
Incorrect acceptance & false holdsCustomer-adjudicated failures accepted and valid submissions blocked.
Time to accepted deliveryElapsed time from task start to acceptance, including repair loops.
Measurement status: paid pilot underway; results not yet reported. The task comparison is scripted and does not supply these measurements.
More agents should not mean more people reconstructing what happened.
Scripted comparison, not a benchmark. Runtime gates are source-inspected; cloud interactions illustrate founder-described capabilities. All sessions, rollouts and records shown are demo data.
08Source-based entry segment
A measured starting pool. An explicit filter.
2,232 U.S. software-publishing firms had 20–499 total employees in 2022. Qualify engineering-team size, agent adoption, budget, and verification pain within that pool.
Assumed share qualifying
1,116 of 2,232 firms qualify
× $24,000 proposed annual contract
$26.8M
Illustrative annual pool
at full penetration of the qualified subset
Expansion thesis
Earn adoption in one team, then add repositories and shared policies. Broader industries, larger enterprises, and international buyers sit beyond this starting segment.
Census SUSB 2022, NAICS 511210. Employee counts include all roles. U.S. entry segment only; other industries and international expansion excluded.
09The problem we focus on
Permission isn’t completion.
An agent can stay within its permissions and still miss a requirement, stop too early or report success without sufficient evidence. Engineering leaders own the consequence.
May it act?
Evaluate permissions, policy and approval before an action executes. LangGuard documents this runtime authority layer.
Has it earned “done”?
Evaluate evidence against phase objectives before work advances. Keep the verdict and the supporting record together.
Enforce the handoff
Combine judge-evaluated phase completion with durable workflow state and signed evidence inside supported agent workflows.
These are overlapping needs, not an empty market. Tiden addresses requirement verification; Open Magi addresses runtime gates and evidence. A judge verdict is not a guarantee of correctness.
10From a team to an organization
One workflow opens a broader account.
The sourced entry pool is a starting segment. Expansion depends on proving repeatable value within customers, then repeating that motion across industries.
One engineering team
Start with one repository, one sponsor and an expensive verification workflow. Prove net value before increasing scope.
Shared delivery policy
Add teams, repositories and release workflows. Centralize acceptance requirements, exception ownership and evidence across the account.
Enterprise software estates
Target internal engineering in healthcare, finance and other industries. Enter when supported integrations and security requirements are proven.
Illustrative scale scenario: 1,000 organizations × $100K annual contract = $100M ARR. Neither buyer count nor enterprise ACV is validated; this is a scale hypothesis, not TAM, valuation or forecast.
11The buyer’s alternatives
A real category. A specific purchase decision.
Teams can buy a full software factory or add governance around existing agents. Legatus starts with the acceptance decision between work produced and work trusted.
8090 + Factory
Both address the broader software lifecycle. Buyers compare how reliably a requirement becomes an accepted change.
Warp + Port
Coordinate agents, context and controls. Legatus must prove a faster path to a repeatable acceptance workflow.
Tiden + Open Magi
Direct overlap in requirements verification, runtime gates and evidence. Evaluate the same failure cases, not feature-count claims.
Inspect the capability matrix and documented alternatives.
Comparison based on vendor documentation reviewed September 29, 2026. No head-to-head superiority claim. Additional technical alternatives: NAEOS, Findry and Nool.
12Our entry strategy
Earn trust in one workflow. Expand from there.
Win one team by reducing the work between an agent claiming completion and a human accepting the change. Keep their supported coding tools and make each decision inspectable.
Control each transition
Shared policy, durable phase checkpoints, judge evaluation and signed proof records support a consistent acceptance process.
Supported Claude Code and Codex paths; coverage is bounded by the integration.
Less supervision
Compare matched tasks against the customer’s current process. Measure human minutes, incorrect acceptance, false holds and cost per accepted change.
Paid pilot: Swordsweeper. Measured outcomes are not yet reported.
Reusable acceptance
Turn validated workflows into versioned requirements, regression cases and reusable policy. Expand as onboarding effort falls.
Next: authenticated review identities and independently verifiable release bundles.
Working mechanisms are source-inspected. Customer advantage and durability of the business remain hypotheses to validate; the full objective-to-production control plane is the vision.
13Pricing and value hypotheses
Land with a pilot. Grow with the team.
Start with one repository and agreed success criteria. Convert to an annual subscription when the team can measure the value.
$5K / 6 weeks
One team, one repository, baseline measurement, supported execution paths, and a joint outcome review.
$24K / year
Team policy, verification records, and support. Customers retain their own agent and model spend.
25 net hours / month
Measure gross hours saved minus added review and intervention. At an assumed $120/hour, 25 net hours create $3K of capacity value against a proposed $2K monthly plan.
Adjust saved time, added review, and hourly cost.
Future offer economics: $5K / 6 weeks; $24K annual plan. Swordsweeper’s actual fee, scope, duration, and conversion terms are not represented by these proposals.
14Current traction / next validation
A paid pilot is underway.
Swordsweeper is a current paid pilot customer. The next step is to document customer outcomes and build a repeatable path to annual contracts.
Swordsweeper
A current customer engagement provides an initial willingness-to-pay signal. Pilot outcomes and annual conversion remain to be demonstrated.
Reliability + value
Compare matched tasks: human minutes, incorrect acceptance, false holds and total cost per accepted change. Agree the baseline and sample before evaluation.
3 paid pilots total
Build from the current pilot through founder-led outreach. Qualify the sponsor, verification burden, and supported workflow before each engagement.
Current fact, proposed measures, and annual conversion.
Swordsweeper paid pilot confirmed September 29, 2026. Fee, duration, scope and measured results await confirmation. Pilot criteria and future acquisition targets are proposals.
15Proposed team after financing
Two hires. Two delivery bottlenecks removed.
Gary leads product, architecture and founder-led sales. The first hires turn founder-dependent implementation and onboarding into repeatable delivery.
Runtime engineer
Own Rust runtime reliability, authenticated review, evidence export and regression testing. Reduce technical key-person dependence.
Unlock: supported workflows that recover reliably and carry verifiable records.
Customer engineer
Own pilot integrations, onboarding, baseline measurement and support. Convert customer requirements into reusable workflows.
Unlock: faster onboarding and customer-reviewed value evidence.
Founder-led selling
Gary retains customer discovery, product direction and early sales. Hire against pilot needs and validated delivery bottlenecks.
Roles are planned, not filled. Sequence depends on financing and pilot demand.
Fund within the existing proposed product/engineering and pilot-delivery allocations; no additional capital assumed. Compensation, start dates and recruiting commitments are not finalized.
16Strategic options / acquisition thesis
Build the business. Create strategic options.
Build a durable, independent business around trusted agent delivery. Strategic value grows with recurring revenue, embedded workflows and proven customer outcomes.
Acceptance inside the SDLC
Potential fit: add verified phase transitions and reusable acceptance policies to an existing development platform.
Evidence from code to production
Potential fit: connect agent actions, policy decisions and release evidence to operational context.
Control across agents and teams
Potential fit: bring governed software delivery and inspectable decisions into a wider enterprise AI platform.
Adjacent acquisition precedentServiceNow acquired Moveworks · 2025
AI assistants and enterprise search.
What creates strategic value
- Retained customer revenue
- Measured delivery advantage
- Reusable policy workflows
- Defensible verification IP
Strategic fit is our inference. These are adjacent acquisition precedents; no buyer interest, exit timing or valuation is represented. Build customer value first.
17Open risks / proposed mitigation
Know the risks. Fund the proof.
Four risks shape the pre-seed plan. Each has a concrete response and evidence to earn before wider deployment or expansion.
01 / OPEN RISK
Reliability & security
Wrong acceptance, bypasses or exposed customer code can destroy trust.
Proposed response
Bound supported paths; test adversarial cases; isolate data; authenticate reviewers.
Evidence to earn
False-accept and false-hold results, access checks and independently verified records.
02 / OPEN RISK
Platform dependence
Harness changes and bundled controls can weaken the wedge.
Proposed response
Version-test adapters; reuse policies across harnesses; compare against native tools.
Evidence to earn
Supported-version coverage and measured advantage over the customer’s current process.
03 / OPEN RISK
Commercial validation
Pilot value, renewals and service margins remain unproven.
Proposed response
Baseline pilot outcomes; measure net time saved; include onboarding and support costs.
Evidence to earn
Paid conversions, recurring usage, customer-approved value and delivery gross margin.
04 / OPEN RISK
Execution & runway
Solo-founder capacity and limited runway can slow delivery.
Proposed response
Hire runtime and customer engineers; document operations; stage spending against evidence.
Evidence to earn
Team-owned delivery and runway reviews at each checkpoint; $150K planned reserve.
Risks remain open. Mitigations and validation evidence are planned unless verified elsewhere. Agree success thresholds with each pilot customer before evaluation.
18Proposed capital-to-evidence plan
Three checkpoints for the next investment case.
Deploy $1.35M across an illustrative three-phase plan, with $150K held in reserve. Each checkpoint links spending to evidence for a future raise.
Technical evidence
Target 3 paid pilots total. Validate required holds, false holds and complete records. Add authenticated review identities and portable verification.
Evidence: customer-reviewed test reports, supported-path coverage, and documented pilot outcomes.
Commercial evidence
Target 5 annual customers at $24K: $120K contracted ARR, including at least 2 pilot conversions. Measure net value and onboarding effort.
Evidence: executed subscriptions, sponsor-approved value assessments, and fully recorded delivery costs.
Repeatable delivery
Target 10 annual customers: $240K contracted ARR. Target ≥80% monthly active accounts for 3 months and ≥70% service gross margin.
Evidence: recurring usage, account expansion, and margin after hosting and direct support labor.
Planning checkpoints, not committed financing tranches. Usage means a supported workflow completed during the month. All customer, ARR, and margin figures are targets; one paid pilot is active today.
Appendix / competitive diligence
Competition deep dive.
Documented overlap, evidence boundaries, and the tests Legatus must win.
| Capability | Legatus | LangGuard | Kontext | PointGuard | NVIDIA | EQTY Lab | Tiden | Open Magi |
|---|---|---|---|---|---|---|---|---|
| Pre-action enforcement | I | D | D | D | D | ? | ? | D |
| Deterministic policy decisions | P | D | ? | ? | D | ? | ? | D |
| Tamper-evident records | I | ? | ? | ? | ? | D | ? | D |
| Judge verdicts on proof chain | I | ? | ? | ? | ? | ? | ? | ? |
| False-completion detection | F | ? | ? | ? | ? | ? | D | D |
| Judge-gated phase advancement | I | ? | ? | ? | ? | ? | ? | ? |
| Anomaly detection | F | ? | ? | D | ? | ? | ? | ? |
| Authenticated independent reviewer | ? | ? | ? | ? | ? | ? | ? | ? |
| Signed policy distribution | F | ? | ? | ? | ? | ? | ? | ? |
| Fleet management | F | D | ? | D | ? | ? | ? | ? |
| Local developer entry | I | ? | D | D | D | ? | D | D |
| Cloud governance plane | F | D | D | D | ? | ? | ? | D |
Definitions and scope differ. Legatus: hooks inspected; Wasm and cloud operation not independently checked here. Its judge can be probabilistic despite fixed verdict labels. Doppler identity does not establish independent reviewer authentication. Tamper-evident records do not necessarily mean a hash chain. False-completion detection includes requirement or evidence checks; it is not universal correctness.
Vendor descriptions establish overlap, not equal capability or production maturity. Open an alternative to see the comparison that matters.
8090Full SDLC platform
Business intent, coordinated execution, oversight and auditability.
Legatus win test: Compare acceptance quality and review effort on the same delivery task.
Read primary sourceWarp FactoriesFactory control plane
Coordinates models and harnesses through specification, implementation, review and verification.
Legatus win test: Compare policy continuity and cost per accepted change across harnesses.
Read primary sourcePortEnterprise platform
Shared engineering context, agent management, workflow orchestration and governance.
Legatus win test: Compare time to install a usable acceptance workflow in one team.
Read primary sourceFactoryAutonomous delivery
Connects intake, planning, execution, validation, shipping and monitoring.
Legatus win test: Compare human intervention and incorrect acceptance on matched tasks.
Read primary sourceTidenRequirements verification
Versioned requirements, linked checks, verification gates and production feedback.
Legatus win test: Compare requirement coverage, false holds and review effort.
Read primary sourceOpen MagiRuntime governance
Policy gates, verifier registry, approvals and an Ed25519-signed evidence ledger. Early beta.
Legatus win test: Compare bypass resistance, authority separation and evidence verification.
Read primary sourceNAEOSEngineering control plane
Connects specifications, policy, controlled execution and verifiable evidence.
Legatus win test: Compare supported workflows and independent verification; maturity unverified.
Read primary sourceFindryGoverned delivery
Engineering graph, policy gates and a hash-chained ledger across the lifecycle.
Legatus win test: Compare impact coverage and portable evidence; signing roadmap differs.
Read primary sourceNoolEngineering state
Goal decomposition, dependency coordination, policy gates and signed implementation evidence.
Legatus win test: Compare continuity across sessions, provenance and onboarding effort.
Read primary sourceLangGuardAction authority
Runtime action policy, approvals and linked decision records.
Legatus win test: Compare delivery acceptance separately from action authorization.
Read primary sourceKontextRuntime authorization
Identity, scoped credentials and action policy; developer-local entry documented.
Legatus win test: Compare supported delivery gates; do not assume enterprise-only adoption.
Read primary sourcePointGuardAgent security
Pre-execution controls, agent identity, cryptographic audit and endpoint protection.
Legatus win test: Compare integrity guarantees and correctness checks on matched tasks.
Read primary sourceNVIDIA OpenShellSecure agent runtime
Sandbox controls across network, filesystem, processes and inference.
Legatus win test: Compare complementary runtime security with delivery acceptance.
Read primary sourceEQTY LabVerifiable runtime
Hardware-rooted oversight and cryptographic attestations of execution.
Legatus win test: Compare evidence trust boundaries; judge-verdict equivalence is unverified.
Read primary sourceLegatus: source-inspected implementation and a founder-confirmed paid pilot. No comparative benchmark results are asserted. Reviewed September 29, 2026.
19Raising $1.5M pre-seed
$1.5M to establish repeatable delivery.
An 18-month plan to prove reliable acceptance, convert pilots into annual contracts, and reach 10 annual customers with measured usage and delivery economics.
Product & engineering55% / $825K
Infrastructure & evaluation15% / $225K
Pilot delivery & sales15% / $225K
Operations5% / $75K
Reserve10% / $150K
Change the runway and inspect each spending allocation.
Proposed allocation: $1.35M operating capital, averaging $75K/month, plus $150K reserve. Next-raise evidence: reliability, paid conversion, recurring usage, and delivery margin. No revenue contribution assumed.