Agent Engineering Wiki

The dominant control structure is still the ReAct loop (reason → act → observe, repeat), and the production lesson is that the loop alone isn't enough — Stripe's financial-compliance agent pairs a ReAct framework with dedicated infrastructure and guardrails to keep multi-step runs on track at production scale, evidence that planning reliability is an architecture problem, not a prompt. Two refinements are emerging on top. First, knowing when to ask vs. proceed: DiscoBench measures clarification-aware deep search, scoring whether an agent recognizes an under-specified goal and asks rather than confidently planning down the wrong path — treating "ask a question" as a first-class planning action. Second, learning to plan from experience rather than re-deriving a plan cold each run: GUI agents that autonomously explore and reuse *hindsight* experience plan repetitive interface tasks better than zero-shot decomposition, and DAIN's dynamic agent-interaction network adapts the collaboration/reasoning structure to the task instead of running a fixed plan. The through-line is that robust planning comes from *structure around the loop* — explicit decomposition, clarification gates, learned priors, and a harness that can re-plan — not from a single cleverer prompt. That the loop itself is now the industry's shared vocabulary for this problem showed up at the AI Engineer World's Fair, where "loops" and "software factories" — production setups that wrap a planning loop in enough infrastructure to run it repeatedly and reliably — were a dominant theme alongside forward-deployed engineering, evidence that planning-as-harness-problem has moved from research framing to mainstream practitioner conversation.

"The loop" is now solidifying into an engineered, reusable artifact rather than a one-off prompt pattern. A provider-agnostic reference implementation built on ports-and-adapters (call model, run tools, feed results back, stop) treats the loop itself as portable infrastructure any OpenAI-compatible backend can plug into, and QUALITY.md proposes an open spec, agent skill, and CLI for grading "loop engineering" quality directly — naming and measuring the harness-quality axis rather than leaving it implicit. Self-improving variants are also emerging: an "autoresearch" pattern has agents iterate on their own task *recipes* across runs, closing a feedback loop over the plan itself rather than just over individual steps, though practitioners are explicit that humans stay central to steering it — the case against one-shot AI design argues skill engineering (iterative, human-curated task specs) beats hoping a single prompt gets the plan right.

The "know when to ask vs. proceed" thread also gains a metacognitive angle: CoMet targets uncertainty estimation directly — decomposing *what kind* of uncertainty a multimodal model has, since "knowing what you don't know" is exactly the signal a planning loop needs to decide whether to ask a clarifying question or charge ahead, extending DiscoBench's clarification-aware benchmark with a mechanism for producing that signal in the first place.

Training is starting to target planning directly, not just the harness around it: OpenAI's Agent RFT fine-tunes reasoning models against reward signals from real tool interactions, using reinforcement learning to solve the credit-assignment problem — which of the many steps in a long trajectory actually caused success or failure — rather than relying entirely on prompting or a hand-built harness to keep the loop on track. AWS SageMaker's multi-turn RL best practices name the same credit-assignment job from the infrastructure side: build a training environment you can trust, run an external evaluation separate from the reward signal, design the reward to actually match the end task, and manage state across turns — the operational checklist underneath "just fine-tune on tool interactions."

Re-planning on failure is also getting a more structured answer than retry-and-hope: rather than a single reflection pass, a multi-hypothesis failure-attribution approach has autonomous research agents generate several candidate explanations for why an experiment failed, weigh them, and re-plan around the most likely cause — treating failure diagnosis itself as a planning step, not just a trigger for blind retry.

The "ask vs. proceed" question is also moving from a benchmark score to a live control signal: Candidly built a per-turn state model (an IO-HMM over signals like message length and semantic alignment) that infers whether a conversation is Engaged, Detailed, Guided, or Disengaging and steers the agent's next-turn behavior accordingly. Closing that loop in production halved disengaging turns (23% → 11%) and shifted traffic toward the high-resolution Engaged state (53% → 64%) — concrete evidence that inferring "is this plan working" mid-episode, not just at the end, is worth the extra model.

Re-planning on failure gets a multi-hypothesis upgrade: rather than one reflection pass, autonomous research agents now generate and weigh several candidate failure explanations before re-planning, and Candidly's production IO-HMM state model shows *live* steering pays off directly — halving disengaging turns (23% → 11%) by inferring conversation state every turn instead of grading the outcome after the fact. AWS's multi-turn RL best-practices guide names the operational checklist (trustworthy training env, external eval, aligned reward, cross-turn state) underneath training approaches like Agent RFT.

Training is starting to target planning directly: OpenAI's Agent RFT fine-tunes reasoning models against tool-interaction reward signals, using RL to solve credit assignment across a long trajectory rather than leaning only on harness engineering. Alongside it, "the loop" is solidifying into reusable, gradeable infrastructure — a provider-agnostic ports-and-adapters reference loop, an open spec for grading loop-engineering quality (QUALITY.md), and self-improving "autoresearch" loops that iterate on their own task recipes — while practitioners push back that human steering (skill engineering) still beats one-shot design. CoMet adds a metacognitive signal (decomposed uncertainty estimation) to the "ask vs. proceed" question DiscoBench already benchmarks.

Bad planning is what turns a capable model into an unreliable agent: it's the source of runaway loops (a cost problem), of confidently wrong work on ambiguous tickets, and of the long-horizon failures that erode trust. The engineering job is to wrap the model's reasoning in a controllable harness — bounded loops, explicit decomposition, clarification checkpoints, and re-planning on failure — and to prove it works with trajectory-level eval rather than hoping a bigger model plans better on its own. Planning sits upstream of orchestration: once you can decompose reliably, the question becomes who executes each step.