As frontier models absorb prompt-level coding tricks through a user-trajectory data flywheel, the durable agent harness converges on two layers models cannot internalize: isolated create-run-delete VM sandboxes and outer-loop context engines that supply institutional intent, living codebase maps, and precomputed KV caches.
As frontier models absorb prompt-level coding tricks through a user-trajectory data flywheel, the durable agent harness converges on two layers models cannot internalize: isolated create-run-delete VM sandboxes and outer-loop context engines that supply institutional intent, living codebase maps, and precomputed KV caches.
OpenHands framed the central paradox of 2026 harness design: coding scaffolding makes harder tasks tractable, user trajectories feed model post-training, and models absorb those habits until harness choice barely moves coding benchmark scores—motivating self-cleaning guardrail shims that watcher bots delete once newer models stop triggering them. What models cannot absorb defines the lasting architecture. First is the execution boundary: Daytona showed that every competing framework reduces to create-run-delete VM primitives and is actively deleting local executors because embedded runtimes still expose host tools and a single allowed `python -c` command compromises the host; Arize and Greptile isolate every run in disposable sandboxes while persisting memory externally. Second is outer-loop context: Linear showed that five-cent intelligence shifts failure from coding capacity to product incoherence unless triage agents assemble customer value, logs, and prior decisions upfront; Greptile replaces grep with a living company wiki; Sonar Vortex feeds structural maps in-loop to cut token spend 36%; Ramp compresses 20–30x month-end context into offline KV-cache cartridges; GEICO routes edge-case HTML worksheets to domain owners; and Basis treats every working-window token as runtime training data. Yet as OpenHands noted in counter-point, entering unmastered domains like autonomous multi-agent colleagues immediately forces teams to re-add heavy prompting, coordinator hierarchies, and guardrails.
“these tools that your agent will call inside that runtime will actually run on your device outside that runtime.”Mislav Ivanda · Give the Model a Computer: The One Thing Every AI Agent Framework Agrees On · 00:03:28
In The Model Ate the Coding Harness. Now What? (00:04:14), Jediah Katz argued that coding harnesses obsolete their own context-layer scaffolding through a data flywheel in which scaffolding makes harder tasks tractable, user trajectories train frontier models natively, and harness choice stops changing coding benchmark scores. In The Model Ate the Coding Harness. Now What? (00:01:22), Jediah Katz argued that an agent harness divides architecturally into a foundational runtime layer—managing core loop execution and durable state persistence—and a context orchestration layer governing model routing, tools, memory, compaction, and behavioral guardrails.
In The Model Ate the Coding Harness. Now What? (00:14:01), Jediah Katz argued that teams can ship deterministic guardrail shims for current model flaws without accumulating permanent codebase debt by logging every shim activation and having an automated watcher agent open a removal PR once newer models stop triggering it. In The Model Ate the Coding Harness. Now What? (00:14:01), Jediah Katz argued that to iterate rapidly without Hyrum's Law lock-in or integration sprawl across Slack messages, CI status alerts, and timers, harnesses should hide internal mechanics from users and expose composable agent primitives like a unified event-subscription tool.
In Give the Model a Computer: The One Thing Every AI Agent Framework Agrees On (00:03:28), Mislav Ivanda argued that host-level command allowlists and approval prompts fail to bound what a compromised model can do because permitting a single interpreter command like `python -c` grants unrestricted socket, filesystem, and code execution access. In Give the Model a Computer: The One Thing Every AI Agent Framework Agrees On (00:07:40), Mislav Ivanda argued that across 2026 releases, agent frameworks such as CrewAI and AG2 are actively deleting legacy local code executors and forcing execution into isolated VM or process sandboxes that can be discarded upon compromise.
In Give the Model a Computer: The One Thing Every AI Agent Framework Agrees On (00:09:18), Mislav Ivanda argued that sandboxes have evolved from short-lived single-snippet executors into full agent runtimes and massively parallel fleets of thousands of concurrent environments powering reinforcement learning post-training, evaluations, and subagents. In Give the Model a Computer: The One Thing Every AI Agent Framework Agrees On (00:00:05), Mislav Ivanda argued that whether exposed as a tool, an external plugin, or a subclassed internal API, competing agent frameworks all reduce sandbox integration to the same three lifecycle primitives: create, execute, and delete.
In What does an AI engineer do? (00:06:32), Aparna Dhinakaran argued that signal decouples ephemeral compute from durable state — tearing down sandboxes, credentials, and repo access after each run while checkpointing step state, detected issues, traces, and learned skills externally across model- and harness-agnostic runners.
“It turns out the harness is actually making itself obsolete through what I call the data flywheel. So in this process, the harness brings harder and harder coding tasks within reach.”Jediah Katz · The Model Ate the Coding Harness. Now What? · 00:04:14
“The runtime is kind of like the foundation of the harness. It controls where the core agent loop runs and how state is stored so that work can resume.”Jediah Katz · The Model Ate the Coding Harness. Now What? · 00:01:22
“So now we have a bot who watches this dashboard, and whenever a shim stops firing, that model or that agent will put up a PR and say this shim can be cleaned up.”Jediah Katz · The Model Ate the Coding Harness. Now What? · 00:14:01
In The Machine that Changed the World, vAI (00:10:56), Tariq Shaukat argued that supplying structural codebase maps and verification guide rails inside the agentic loop via Sonar Vortex reduces token spend by 36% and cuts commit-time issues by 92% on large, long-running tasks. In Your Coding Agent Has a Context Problem (00:02:20), Tom Moor argued that as intelligence that cost $100 three years ago drops to five cents and agents run unattended for hours or days, context can no longer be hand-fed interactively and must be orchestrated by outer-loop systems.
In Your Coding Agent Has a Context Problem (00:06:18), Tom Moor argued that cheap, abundant code shifts the core failure mode to product incoherence because agents reading source code alone miss user intent, prior design trade-offs, parallel org work, decision ownership, and recent revert risks. In Your Coding Agent Has a Context Problem (00:08:47), Tom Moor argued that linear's triage agent deduplicates multi-channel intake and enriches issues via internal MCPs with customer value, request frequency, Datadog logs, stack traces, live bug reproductions, and prior related decisions before coding starts.
In How to build an AI code review agent: from the creators of Greptile (00:10:45), Daksh Gupta argued that replacing imprecise, token-heavy grep searches with a continuously maintained wiki that synthesizes codebase architecture, production telemetry, docs, and tickets gives review agents surgical tribal context.
“And what we see when we provide agents in long-running, large, complex databases with long-running tasks, with this information is a 36 percent reduction in token cost because you're putting it in white claims.”Tariq Shaukat · The Machine that Changed the World, vAI · 00:10:56
“I think one of the main reasons is that the context kind of moves out of that inner loop and largely into the outer loop.”Tom Moor · Your Coding Agent Has a Context Problem · 00:02:20
“Implementation is not the bottleneck now. The failure mode here isn't an inability to build. The failure mode is like an inability to maintain coherence, to have a product that makes sense as a whole.”Tom Moor · Your Coding Agent Has a Context Problem · 00:06:18
In Process Supervision for Agents That Do Long-Horizon Work (00:09:16), Mitchell Troyanovsky argued that in long-horizon agent harnesses, behavior emerges jointly from system prompts, progressive disclosures, tool descriptions, and model weights, so every token entering the working window must be engineered as runtime training data. In Inside the Fable War Room (00:11:18), Ford Prior argued that at each milestone, the war room generates interactive HTML worksheets and editable architecture diagrams that route sharp edge-case questions to domain owners and ingest human edits back into agent context.
In Chasing capabilities while waiting for the frontier (00:05:09), Alexander Shevchenko argued that offline KV-cache cartridges compress 20x to 30x more predictable context—such as ledgers and aging reports prepared overnight before month-end close—with lower engineering lift and broader generalizability than post-training.
“really think of the context that agent is going to see as it's working as runtime training data. The word prompt, I think, trivializes it a bit, right?”Mitchell Troyanovsky · Process Supervision for Agents That Do Long-Horizon Work · 00:09:16
“It was like, wow, that's actually a great question that we've never thought about identifying an edge case either in the business with technology.”Ford Prior · Inside the Fable War Room · 00:11:18
“cartridges are really interesting family of techniques where you're able to compress a lot of context, up to 20 to 30 times more context, into the KB cache offline and serve it for your agent.”Alexander Shevchenko · Chasing capabilities while waiting for the frontier · 00:05:09
No operational pattern at AI Engineer NYC went unqualified. Across the corpus, 1 counter-claim pushed back on naive versions of this shift—highlighting where judges introduce variance, where caches break, or where human judgment stays irreducible.
“So Rockbot has a new kind of harness that operates not based off of one-off chats, but long-lived coordinator bots that orchestrate a fleet of worker agents.”Jediah Katz · The Model Ate the Coding Harness. Now What? · 00:06:55
Will frontier models absorb the agent harness, or must builders keep engineering scaffolding and process around them?
Competing agent frameworks that disagree on nearly every abstraction all converge on isolating code execution inside a separate VM sandbox via create-run-delete primitives, making the execution boundary the primary architectural commitment to lock in before choosing a swappable framework.
Building a production-grade AI code review agent requires moving beyond naive diff prompting to combine historical incident eval datasets, cross-family author-reviewer model inversion, a living codebase wiki instead of raw grep, dependency-rich sandbox UI testing, and per-subtask model routing.
Because coding agents accelerated inner-loop implementation while leaving enterprise legacy modernization bottlenecked on cross-functional alignment and domain context, GEICO built a multiplayer AI-native war room steered by Markdown FleetKit protocols, human-in-the-loop worksheets, and deterministic nine-step gates to ship legacy claims APIs in a single day.
The AI engineer's role has shifted from manually building and reviewing individual features to architecting closed-loop agent factories — where agent-first telemetry triggers isolated, durable sandbox runs, LLM judges are meta-evaluated against human judgment, and both production and coding agents are traced end-to-end.
Even as frontier models rapidly drive down unit token costs, Ramp argues it still pays to invest in applied research below the API layer—combining frontier orchestrators with post-trained specialist models, offline KV-cache cartridges, and cheap open-weight agent swarms to cut latency and cost today.
Enterprise agents achieve production adoption not through open-ended autonomy or outcome-only evals, but by encoding opinionated workflows as behavior specs that grade ground-truth-free trajectories and enable targeted debugging via trajectory forking and subagent simulation.
Because abundant AI-generated code overwhelms human reviewers, compounds multi-step errors, and defeats inconsistent LLM judges, enterprise software factories must engineer shipping confidence multiplicatively through clear intent, in-loop guide rails, hybrid deterministic-generative verification, and auditable evidence.
As frontier models natively absorb coding-specific context scaffolding through a user-driven data flywheel, the agent harness shifts from adding raw coding intelligence to unlocking unmastered multi-agent domains—like autonomous knowledge workers and software factories—while owning durable cloud runtimes, scoped security, token efficiency, and self-cleaning primitives that models cannot absorb.
As inference costs plummet and coding agents run unattended in the outer loop, engineering's primary failure mode shifts from implementation capacity to product incoherence, requiring automated pipelines that thread customer intent, product history, and ownership through triage, implementation, review, and daily feedback distillation.