Product.ai / Join / AI Systems
AI Systems

Where is AI Systems
work heading at frontier firms
in 2026?

Models commoditized. Context engineering, evaluation architecture, and verification certainty became the new moats. The 10x AI Systems operator architects the oracle that decides what is true — not the model that generates the candidate. Below: open projects, the physics of what is changing, and roles we are hiring against.


Open challenges

Frontier AI systems work challenges we are working on right now.

Real, paid 1-3 week engagements with the Product.ai team. Each one is a problem we are working on at the frontier of AI systems work — and the kind of work we hire against. Pick one that pulls at you and apply.

Agent Confidence-Signaling Component — how Alloy speaks confidence to other AI agents and to the human in the loop
2 weeks · Agent commerce + Truth Graph · Consequential Open to Alpha Team Draft
Design and ship a confidence-signaling component for Alloy — Product.ai's local-first agentic workbench.
Production Eval Harness — error-analysis-first failure taxonomy on Product.ai chat or Alloy
2 weeks · Product.ai + Truth Graph · Consequential Open to Alpha Team
Build a production eval harness for one Product.ai surface (chat or Alloy).
Verification Ladder + Back-Pressure Hooks — close the deterministic gate on one Cortex or AIOS agent workflow
2 weeks · Engineering + Product.ai · Applied Open to Alpha Team
Audit one agentic workflow inside Cortex or AIOS — a cron-triggered skill, an autonomous pipeline, a sub-agent fan-out — and ship the deterministic verification ladder on it.
Cost-Split Multi-Model Routing — make Haiku, Sonnet, and Opus do their actual jobs across one cron pipeline
2 weeks · Engineering + Product.ai · Applied Open to Alpha Team
Pick one Cortex or AIOS cron pipeline that currently calls Opus 4.7 on every step and refactor it to cost-split routing — Haiku 4.5 or Sonnet 4.6 for routine sub-steps, Opus 4.7 only on escalation, advisory, or synthesis steps.
External Truth Anchor — /feedback channel for Product.ai chat or Alloy that reaches engineering Slack within 60 seconds
1 week · Product.ai + Engineering · Foundational Open to Alpha Team
Ship an in-product `/feedback` channel for one Product.ai surface (chat or Alloy).
MCP Throughput Refactor — diagnose and fix the protocol-overhead trap on a Product.ai backend MCP path
2 weeks · Agent commerce + Product.ai · Consequential Open to Alpha Team Draft
Audit one Product.ai MCP server for protocol overhead, request shape, and unit economics at production load.
Sub-Agent Orchestration with Worktree Isolation — refactor one Cortex or AIOS workflow into the engineer-plus-agent-fleet pattern
2 weeks · Engineering + Cortex · Consequential Open to Alpha Team Draft
Pick one Cortex or AIOS workflow currently running sequentially (or with naive sub-agent dispatch) and refactor it into the engineer-plus-agent-fleet pattern.
LLMAdapter Constitutional Implementation — thin sovereignty across one Product.ai backend surface
2 weeks · Engineering + Product.ai · Consequential Open to Alpha Team
Define and ship the `LLMAdapter` interface as Product.ai's constitutional pattern for talking to model providers.
Per-PR Cost Ledger + Agent-Policy Metadata — Day 1 substrate for safe agent deployment across the Product.ai repos
2 weeks · Engineering + Cortex · Applied Open to Alpha Team
Build the two artifacts a backend platform should ship before opening agent deployment to the team: (1) per-module agent-policy metadata declaring agent-allowed / agent-allowed-with-mock / agent-forbidden across each Product.ai backend r...
Trace-to-Regression-Test Pipeline — close the user-feedback loop with a 24-hour SLA
2 weeks · Engineering + Product.ai · Consequential Open to Alpha Team
Build the production pipeline that converts user-flagged failures into regression tests against a 24-hour SLA.
External MCP Server v1 — Product.ai's commerce-knowledge query endpoints exposed for AI agents
3 weeks · Agent commerce + Product.ai · Consequential Open to Alpha Team Draft
Design and ship Product.ai's external-facing MCP server v1 — the surface AI agents (Claude, ChatGPT, Gemini agents, custom agents) call to query verified commerce knowledge.
Production-Trace Failure Taxonomy — review 100 traces, build the eval substrate from real failures
2 weeks · Product.ai + SimplyCodes · Consequential Open to Alpha Team
Pick one Product.ai surface (chat, Alloy, or SimplyCodes' code-verification fleet).
Anthropic Postmortem Calibration + Product.ai Eval-Bypass Mitigation — diagnose one equivalent risk and ship the gate
2 weeks · Product.ai + Engineering · Consequential Open to Alpha Team Draft
Read the Anthropic April 23, 2026 Claude Code postmortem end-to-end.
Agent Commerce PRD v1 — write the spec for one external-facing MCP capability with eval criteria scoped correctly
2 weeks · Agent commerce + Product.ai · Consequential Open to Alpha Team Draft
Write the v1 PRD for one external-facing MCP capability — the surface other AI agents will call to query Product.ai's verified commerce knowledge.
SimplyCodes Working-Code-Rate Lift — drive measurable improvement on the verification accuracy metric
2 weeks · SimplyCodes + Revenue · Applied Open to Alpha Team
Pick one mechanism limiting SimplyCodes' working-code rate (currently 67%) or the surrounding 96% availability metric.
Cross-Surface Eval Taxonomy — design the measurement system from scratch on one Product.ai surface, prove substrate-builder phenotype
2 weeks · Product.ai + Truth Graph · Consequential Open to Alpha Team
Pick one Product.ai surface (Alloy, Cortex memory, SimplyCodes code-verification, or Product.ai chat).
Layer 4 Human-Decision Compliance Instrumentation — measure adoption-decay and override patterns on one Product.ai surface
3 weeks · Product.ai + Truth Graph · Consequential Open to Alpha Team Draft
Build Layer 4 instrumentation on one Product.ai surface — the surface where the agent's recommendation reaches a human and the human acts on it, overrides it, or ignores it.
Procedural Integrity Audit — measure the corrupt-success tax on one Product.ai agentic workflow
2 weeks · Product.ai + Truth Graph · Applied Open to Alpha Team Draft
Take one Product.ai agentic workflow (Alloy, ARC application-layer if accessible, signal-step-executor, an AIOS skill that fans out sub-agents).
SimplyCodes Conversion Attribution Model — replace heuristics with a defensible attribution stack
2 weeks · SimplyCodes + Revenue · Applied Open to Alpha Team Draft
Replace SimplyCodes' current conversion attribution heuristics with a defensible attribution model.
Agent-API Recommendation Attribution (Layer 2) — instrument the recommendation-decision moment
3 weeks · Product.ai + Agent commerce · Consequential Open to Alpha Team Draft
Build Layer 2 instrumentation — the API gateway agent attribution layer that captures which AI agents (Claude Code, Cursor, ChatGPT, Gemini, custom agents) recommend Product.ai or SimplyCodes APIs in their reasoning loops.
AIOS Skill Creation End-to-End — build one new skill through /skill-create, register it properly, prove the skill compounds
1 week · Engineering + Cortex · Foundational Open to Alpha Team
Build one new AIOS or Cortex skill end-to-end.
CEO-Bottleneck Agent Workflow — automate one workflow currently bottlenecked on Michael's bandwidth
2 weeks · Cortex + Engineering · Applied Open to Alpha Team
Identify one workflow currently bottlenecked on Michael's bandwidth — Slack inbox triage, Gmail processing, meeting prep, decision-support synthesis, signal classification, claim curation, or another concrete pattern.
Builder Trial End-to-End — pick one Product.ai problem and ship the agent-based solution
2 weeks · Product.ai + SimplyCodes · Applied Open to Alpha Team
Pick one concrete Product.ai problem — a SimplyCodes operations workflow, a Cortex curation pattern, an AIOS team-coordination gap, an Alloy edge case.
Cron Job Health Audit + Consolidation — survey ops/cron, ship the consolidation, instrument the health surface
2 weeks · Engineering + Cortex · Applied Open to Alpha Team
Audit the cron jobs in `ops/cron/` — every active job, every silently-failed job, every duplicate, every overlapping ownership.

Discipline physics

Where AI systems work is heading at frontier firms.

Last updated

In May 2026, AI Systems work at frontier firms (Anthropic, OpenAI, Cohere, Mistral, Inflection, Adept, Scale, Anysphere) has bifurcated cleanly. Three job families — Research Engineer, ML/Eval Engineer, AI-Native Data Engineer — share tooling but not craft. The shared physics: <strong>the model is a commodity; the context, the eval, and the verification protocol are the moat.</strong> Senior practitioners do not measure themselves by accuracy at training time. They measure by what holds up under adversarial paraphrase, distribution shift, and closed-loop deception. The 10x operator is the one who treats the evaluation framework as a shippable executable contract, not a slide.

Three families inside one umbrella

The "AI Systems" label collapses three structurally distinct families. Research Engineers implement and run experiments at the model surface — pre-training, fine-tuning, RLHF, alignment. They're found at frontier labs (Anthropic, OpenAI), not at every product company. ML/Eval Engineers are the largest category — they build production inference, evaluation, monitoring, and deterministic-gate infrastructure around models. AI-Native Data Engineers architect the context substrate: structured commerce data, knowledge graphs, retrieval indices, embedding stores, the canonical truth that models read from.

These families share a stack (Python, PyTorch/JAX, vector DBs, LLM SDKs) but optimize for different objectives. Research Engineers optimize for capability. ML/Eval Engineers optimize for verified production behavior. Data Engineers optimize for context fidelity. A 10x Research Engineer who tries to ship product evals will under-invest in adversarial coverage; a 10x Data Engineer who tries to write a fine-tune will under-invest in distribution-shift defense. Hire to the family, not the umbrella.

The eval is the product

The legacy ML deliverable was a trained model — a `.pkl` artifact you handed to the engineering team. In 2026 the deliverable is an executable evaluation contract. The eval framework is the specification of what "good" means. Anthropic's public eval philosophy, OpenAI's evals repo, and Scale's Spellbook all converge on this: the team that ships the better eval ships the better product, even if the underlying model is the same. The eval is what calibrates trust, exposes failure modes before users find them, and gates production deployments.

The senior practitioner spends more time designing the rubric than running the experiment. They distinguish glass-evaluator structural checks (path placement, schema validity — bright-line answers, regex works) from stone-evaluator semantic checks (is this content sovereign? does this answer cite real evidence? — meaning is not pattern-matchable). They build regression test corpora with paraphrase variants and prompt-injection attempts. The frontier moat is not "we have a better model." It's "we have a better oracle that decides what counts as a correct model output."

Verification certainty, not predictive accuracy

The legacy DS metric — accuracy on a held-out test set — has eroded as a meaningful signal. Verification certainty replaced it: the probability that an output is correct AND the system knows whether it is correct. A model with 92% accuracy and zero calibrated uncertainty is operationally worse than a model with 87% accuracy and reliable I-don't-know detection. The Trust Paradox documented at Product.ai (74% rate trust 4-5/5; 93% verify before acting) is not a survey artifact — it's the user's correct response to systems that don't signal their own confidence.

The 10x AI Systems operator instruments behavioral verification: citation-click rates, override rates, escalation rates, retry rates. These are the load-bearing trust signals. They beat survey-based scoring in every controlled study. The operator who can articulate which of their production systems systematically over-state their own confidence is higher-signal than the one who can't. Calibrated uncertainty is the new moat. Anyone who treats "model accuracy" as the primary metric is a generation behind.

The agent-economy infrastructure layer

Frontier AI Systems work in 2026 is increasingly about making models talk to other models reliably. Agent-to-agent protocols (MCP, OpenAI Agents SDK, Anthropic's Computer Use), tool-call routing, multi-agent orchestration, and verified-truth APIs are the fastest-growing area. Product.ai's thesis — that verified commerce knowledge is the agent-economy infrastructure layer — is one expression of this physics. Agents need ground truth that survives paraphrase and adversarial input; the firm that ships that infrastructure first wins the coordination layer.

This is why MCP server engineers, agent-tool integrators, and verified-knowledge graph architects have become some of the highest-leverage seats at frontier firms. The work is unglamorous from the outside (defining a tool schema, writing an authentication flow, designing a retry protocol) and load-bearing from the inside (every downstream agent that calls your tool inherits your design choices). 10x operators in this layer think about adversarial use, partial-failure modes, and protocol evolution — not just happy-path correctness.

What the industry got wrong

The "MLOps" framing of 2021-2024 systematically under-invested in evaluation and over-invested in deployment automation. The result was a generation of teams who could ship a model in five minutes but couldn't tell you whether the model was getting worse over time. Anthropic's and OpenAI's 2025-2026 postmortems both cite "eval coverage gaps" as the dominant root cause of production incidents — not deployment failures, not infrastructure issues, but failure to measure what matters.

The companion mistake: treating the model as the bottleneck. In every observable case, the bottleneck is upstream — context quality, retrieval relevance, eval calibration, prompt design. Teams that swap GPT-4 for Claude or vice versa rarely see the gain they expect. Teams that improve their context substrate, their retrieval, or their eval rubric routinely see 2-5x quality improvements with the same model. The frontier moat is below the model, not at it. Practitioners who agree with this framing voice it specifically — context-engineering, retrieval-architecture, eval-design — not in slogans about "AI capability." The frontier is in the dissent, not the consensus.


Open roles

Full-time AI systems work roles open right now.

Most of our best people came through projects, not interviews. If a project pulls at you and the trial goes well, the role conversation follows.

AI Chief of Staff
Hiring · $300k - $400k · HQ - Los Angeles
AI Engineer
Hiring · $250k - $500k · HQ - Los Angeles
Senior AI Engineer, Evals
Hiring · $375k - $450k · HQ - Los Angeles

Other disciplines

Working at the edge of an adjacent discipline?


Apply

If a AI systems work challenge above is the kind of work you want to be doing this month, send a screen.

Twelve-minute Hireflix video, async. Then a 30-60 minute chemistry call. Then a paid 1-3 week project alongside the team. We will know within a week whether to move forward.