Here is the question every serious buyer asks before trusting a solo, AI-native practice with their vehicle, their aircraft, or their production line: can one person and an agent fleet be trusted with a safety-critical system? It is the right question. This page answers it head-on — and turns the answer into the central reason to work here.
The founder signs; agents assist — never the inverse.
Named-human accountability is the pitch — not “watch the fleet build it.” One senior engineer owns the safety case end-to-end, force-multiplied by audited AI agents: fewer hands, more accountability, full traceability.
Last reviewed: June 2026
Safe by design, in this practice, means that accountability for a safety-critical system is concentrated in one named senior engineer who owns the safety case end-to-end, and is force-multiplied — never replaced — by audited AI agents. The agents draft, search, and cross-check at a scale one person could not reach unaided; the engineer reviews, reworks, and signs every artefact before it enters the safety argument. Combined with a 90-day ship/pivot/kill checkpoint on every initiative and explicit co-delivery when scope exceeds one person, the result is a model with more accountability on the critical path than a conventional team of rotating associates — not less.
A solo, AI-native firm doing safety-critical physical AI faces one dominant objection, and pretending otherwise would be dishonest: how can one person plus an agent fleet be trusted with my vehicle, my aircraft, or my production line? If the work were a marketing site or an internal dashboard, the question would be academic. It is not. When the system can hurt someone, the buyer is right to demand a clear, named answer to “who is responsible?”
The wrong way to answer is to point at the technology and say “watch the fleet build it” — to make AI autonomy the selling point. That answer fails precisely where it matters: it leaves no human answerable for the result. In a safety-critical context, that is not innovation; it is an abdication.
The right way to answer is to make accountability the selling point. The differentiator is not how much the agents do; it is who is responsible for what they produce. That is the whole argument of this page, and it can be compressed to one sentence: the founder signs; agents assist — never the inverse.
The common model
Most firms put junior associates on your safety case.
Here
Here, one senior engineer owns it end-to-end — 17+ years across safety-relevant connected-vehicle, network/video, and EV-charging platforms.
The common model
More hands on a safety case dilutes accountability — when something is wrong, no single person is answerable for it.
Here
Fewer hands, more accountability. One name on the assurance case, full traceability from requirement to evidence.
The common model
"Watch the fleet build it" treats AI autonomy as the selling point — and leaves no human answerable for the result.
Here
The founder signs; agents assist — never the inverse. Named-human accountability is the pitch.
Most firms put junior associates on your safety case. Here, one senior engineer owns it end-to-end. That is not a slogan — it is the operating model, and it is what makes a one-person practice a stronger accountability proposition for safety-critical work, not a weaker one.
The principle: the founder signs; agents assist — never the inverse. Fewer hands, more accountability, full traceability.
A single named senior engineer — 17+ years across Cisco (network and video platforms serving 100M+ users), Renault-Nissan-Mitsubishi (a connected-vehicle platform serving 4M+ users across 39 countries), and ABB E-mobility (EV charging infrastructure) — owns the safety case end-to-end. Not a rotating bench, not a pyramid of associates: one person who is answerable for the argument from hazard analysis through to the evidence that closes it.
That engineer is force-multiplied by AI agents — but the agents are audited, not autonomous over the outcome. They draft, search, cross-check, and surface inconsistencies at a scale one person could not reach unaided. Every artefact an agent touches is reviewed and owned by the named engineer before it enters the safety case. Fewer hands, more accountability, full traceability.
Safety-critical evidence cultures — ISO 26262, DO-178C, IEC 61508 — demand a named responsible engineer behind every claim. An AI-native practice does not weaken that; it strengthens it. There is exactly one author to trace any decision back to, and a complete record of how each piece of evidence was produced and reviewed.
Force-multiplication is real and it is significant. But there is a bright line, and it is the foundation of the whole trust argument: agents assist, the founder builds and signs. The agents accelerate the work; the named engineer remains answerable for every artefact that enters the safety case.
The bright line: an agent's output is always an input to be reviewed and owned by the named engineer — never the signed artefact. A passing automated check is necessary, never sufficient, for the safety argument.
Agents assist
Draft candidate hazard analyses, requirement decompositions, and assurance-case fragments for the engineer to review and rework.
The founder owns & signs
The engineer owns the hazard analysis. An agent's draft is an input, never the signed artefact.
Agents assist
Search standards, prior evidence, and the technical record to surface gaps, contradictions, and missing links faster than manual review.
The founder owns & signs
The engineer decides what the gap means and how to close it. Agents flag; the human adjudicates.
Agents assist
Cross-check traceability — that every requirement maps to a verification, and every claim to evidence — across thousands of links.
The founder owns & signs
The engineer signs off the trace. A passing automated check is necessary, never sufficient, for the safety argument.
Agents assist
Maintain the audit trail: who produced each artefact, when, from what inputs, and which review closed it.
The founder owns & signs
The audit trail records that a human reviewed and owned each step. The agents document the work; they do not certify it.
The fastest way to judge whether named-human accountability is real is to ask for references and put hard questions to the person who would own your safety case. Book a call and do exactly that.
Accountability is not only about who owns the work — it is about being honest when the work is not delivering. Every initiative carries a 90-day checkpoint with three possible outcomes: ship, pivot, or kill. Clear graduation criteria are set up front, and monitoring is in place from day one.
“Permanent pilot” is not in my vocabulary. An initiative that cannot demonstrate it has met its criteria at the checkpoint does not quietly continue consuming budget and attention.
Every initiative starts with explicit, written graduation criteria — what "working" means in measurable terms — agreed before any work begins. There is no ambiguity about what success looks like at the checkpoint, because it was defined when expectations were clearest.
Instrumentation and review cadence are in place from the first day, not bolted on at the end. Progress against the graduation criteria is visible throughout, so the checkpoint decision is informed by evidence accumulated over the full window — not a snapshot taken under deadline pressure.
At the 90-day checkpoint, the initiative graduates to production (ship), changes direction against what the evidence showed (pivot), or stops (kill). "Permanent pilot" is not in my vocabulary: an initiative that cannot demonstrate it has met its criteria does not quietly continue consuming budget and attention.
The honest counterpart to named-human accountability is an explicit model for when scope exceeds one person. A one-person practice must never be a single point of failure on the critical path of a vehicle, an aircraft, or a production line. When the work is bigger than one accountable owner should carry alone, the answer is to partner and co-deliver — not to over-promise.
Before any engagement is accepted, its scope is measured against what one senior engineer plus an audited agent fleet can deliver to a safety-critical standard. If the work exceeds that, it is said plainly — and a co-delivery path is proposed rather than the work being over-promised.
When an initiative is larger than one person should responsibly own alone, the model is to partner and co-deliver with the right specialists or your in-house team — not to stretch a solo practice past the point where accountability holds. "Solo" describes the accountable owner, not a refusal to bring in additional capacity when the safety case requires it.
A one-person practice must never be a single point of failure for a vehicle, an aircraft, or a production line. Continuity arrangements, documented hand-over, and explicit co-delivery for safety-critical scope are part of the engagement design — so the dependency is on a traceable process and named partners, not on one individual's availability.
The accountability model is not an aesthetic preference — it is what safety-critical evidence cultures require. ISO 26262, DO-178C, and IEC 61508 are built on the premise that a named, responsible engineer stands behind every safety claim, with traceable evidence linking requirements to verification. An AI-native practice honours that premise; it does not route around it.
Road Vehicles — Functional Safety
The automotive functional-safety standard. Its safety lifecycle — from hazard analysis and risk assessment through ASIL decomposition to verification evidence — is built on the premise that every safety requirement has a named responsible engineer behind it. The accountability model here is shaped directly by that demand.
Software Considerations in Airborne Systems and Equipment Certification
The standard governing airborne software. It requires rigorous, reviewable evidence linking objectives to verification at each Design Assurance Level (DAL). Its culture of independent review and traceable authorship is exactly the discipline an AI-native practice must honour, not bypass.
Functional Safety of E/E/PE Safety-Related Systems
The foundational functional-safety standard, defining Safety Integrity Levels (SIL 1–4) and the systematic lifecycle for safety-related systems. Its sector derivatives govern machinery and process safety. Like its peers, it expects a named, accountable engineer behind the safety argument.
The following are public, verifiable credentials. None of them is a safety or security certification, and none is represented as one.
The founder is a member of the Forbes Technology Council — an invitation-only community for senior technology leaders. This is a public, verifiable credential. It is not a safety certification and is not represented as one.
The founder serves as a French Government AI Ambassador under the Mission French Tech programme. This is a public-facing role in AI ecosystem development. It speaks to standing and reach in the field; it is not a functional-safety accreditation.
This practice does not hold, and does not claim, ISO 26262, DO-178C, IEC 61508, SOC, or any other safety or security certification. Where a deployment requires certified assessment or sign-off, that is performed by the appropriate accredited body or partner. The differentiator offered here is named-human accountability and traceable engineering rigour — not a certificate this practice does not possess.
Trust in safety-critical work rests on accountability and traceability, not headcount. The model here is that one senior engineer — with 17+ years across connected-vehicle, network/video, and EV-charging platforms — owns the safety case end-to-end and is personally answerable for it. AI agents are audited force-multipliers: they draft, search, and cross-check at scale, and every artefact they touch is reviewed and owned by the named engineer before it enters the safety case. The founder signs; agents assist — never the inverse. Compared with a model where junior associates rotate through your safety case, this puts more accountability on the critical path, not less.
The agents draft candidate hazard analyses and assurance-case fragments, search standards and prior evidence to surface gaps, cross-check traceability across thousands of requirement-to-evidence links, and maintain the audit trail. The bright line: agents assist, the founder builds and signs. An agent's output is always an input to be reviewed and reworked by the named engineer — never the signed artefact. A passing automated check is necessary but never sufficient for the safety argument; a human adjudicates every claim.
Every initiative carries a 90-day checkpoint with one of three outcomes: ship, pivot, or kill. Graduation criteria are defined in writing on day zero, and monitoring is in place from the first day so the checkpoint decision is evidence-based. "Permanent pilot" is not in my vocabulary: an initiative that cannot demonstrate it has met its criteria at the checkpoint does not quietly continue. This protects your budget and forces honesty about whether the work is delivering.
It must never be, and the engagement is designed so it is not. Scope is assessed honestly before commitment; when an initiative is larger than one person should responsibly own, the model is to partner and co-deliver rather than over-promise. Continuity arrangements, documented hand-over, and explicit co-delivery for safety-critical scope mean the dependency is on a traceable process and named partners — not on one individual's availability. "Solo" describes the accountable owner, not a refusal to bring in capacity when the safety case demands it.
No. This practice does not hold, and does not claim, any safety or security certification. The work is shaped by the culture of those standards — named responsible engineers, traceable evidence, independent review — and where a deployment requires certified assessment or formal sign-off, that is performed by the appropriate accredited body or partner. The differentiator offered is accountability and engineering rigour, stated plainly, not a certificate this practice does not possess.
Because in safety-critical work, diffuse responsibility is itself a hazard. When many hands touch a safety case and no single person is answerable, gaps fall between roles and nobody owns the whole argument. A named engineer who owns the case end-to-end, force-multiplied by audited agents, gives you fewer hands, more accountability, and full traceability — exactly the properties ISO 26262, DO-178C, and IEC 61508 cultures are built to demand.
If you are evaluating who to trust with a safety-critical system, the right test is simple: ask who is personally answerable for the safety case, and ask to speak to people they have delivered for. The founder signs; agents assist — never the inverse. Start with a conversation, and ask the hard questions.
Founder & AI Strategy Lead
Mohammed Cherifi is the founder of Hyperion Consulting, with 17+ years across Cisco (network and video platforms), Renault-Nissan-Mitsubishi (connected-vehicle platform across 39 countries), and ABB E-mobility (EV charging infrastructure). He is a Forbes Technology Council member and a French Government AI Ambassador (Mission French Tech), and owns the safety case end-to-end on every engagement.
Functional-safety evidence mapped to ISO 26262, DO-178C, and IEC 61508
Building the safety case for machine learning at the edge under ISO 26262 / IEC 62443
From simulation to production-grade autonomy, with safety architecture that holds