Thinking of ACE? We Can Do It with Fewer Tokens

ALTK-Evolve and ACE both let an agent learn from its own trajectories. The difference is what they do with what they learn — and that decides the token bill.

Give an LLM agent a realistic multi-step task — split a bill, find a song, reconcile an order across nine simulated apps — and when it fails, it usually isn't for lack of knowledge. It mis-paginates an API, resolves the wrong person, or returns a value when none was asked for. The model knows the APIs; what it hasn't internalized is how to use them reliably . That's learnable from the agent's own history.

Two recent systems do exactly this, on the same kind of agent: ACE (Agentic Context Engineering) and our ALTK-Evolve ( introduced here ). Both are a form of agentic memory — turning an agent's past trajectories into reusable lessons and feeding them back at inference time, no weight updates, no human labels. They even agree on the hard part. Where they part ways is delivery .

A note on words, because the two systems name things differently: we'll call the raw thing an agent learns a lesson. ACE organizes its lessons into one comprehensive, evolving playbook; we consolidate ours into individually retrievable guidelines. Same lessons, two containers.

What we agree on

Both systems refuse to compress.

ACE names the failure modes precisely: brevity bias — optimization collapsing toward short, generic instructions — and context collapse — a model asked to rewrite its whole context each step summarizing the detail away. Its answer is to keep a rich, itemized playbook, with a helpful/harmful counter on every bullet, and let the model distill relevance at read time.

We reach the same conclusion from the other direction. Every distinct guideline keeps a support count — how many independent episodes produced it — and we never summarize the store down to a handful of rules. A lesson five different tasks discovered is a different object from one that appeared once, and both are worth keeping.

So on the core question — should you compress an agent's hard-won lessons into a tidy summary? — ACE and ALTK-Evolve give the same answer: no. Count them, don't collapse them. ACE's per-bullet counters and our support counts are two spellings of the same idea.

Where we differ

Two places: how the memory is built, and how it's delivered — and it's the delivery difference that shows up in the token bill.

Consolidation (how the store is built). ACE grows one playbook through a Generator → Reflector → Curator loop, applying incremental delta updates and de-duplicating by embedding. We cluster near-duplicate lessons and merge within a cluster, support-conserving — when several lessons merge, the survivor inherits their combined count, so the store shrinks without losing the record of how much experience backs each guideline. We also extract typed guidelines — strategy, recovery, and optimization — with causal attribution and provenance back to the source trajectory, and at subtask granularity, so a lesson learned on one app can transfer to another.

Delivery (what reaches the model at inference). This is the one that drives the numbers. ACE injects the comprehensive playbook on every step, the same way regardless of model or task. We treat delivery as a dial, not a constant: a small fixed core of high-support guidelines, extended per task with a handful selected for the task at hand (cosine or LLM-guided, priority-weighted) — or, when a model has the headroom to use it, the full consolidated set. The same lessons are available to both agents; the difference is that ACE always sends all of them, and we send however many a given model can actually use.

Why it matters

On AppWorld, with the same base ReAct agent, running both systems in-house:

Model

TGC / SGC Tokens/task

DeepSeek-V3.2 ACE 80.4 / 73.2 634K

ALTK-Evolve 89.3 / 80.4 263K

gpt-oss-120b ACE 54.8 / 35.7 777K

ALTK-Evolve 56.0 / 37.5 116K

On the strong model we're better on both metrics at ~40% of ACE's inference cost. On the weak model we edge ACE 56.0 to 54.8 — close enough that we call it a tie on accuracy (a repeat run of ours landed at 54.8, matching ACE almost exactly, which is within this benchmark's run-to-run noise) — at about one-seventh the cost.

A fair word on cost: ACE's own efficiency story is about building its context cheaply. Ours is on a different axis — serving it. Retrieving a few guidelines per task instead of injecting the whole playbook on every step is where the tokens go, and it's the direct consequence of the delivery difference above.

Where does the accuracy come from? The by-difficulty breakdown tells two different stories: