Match the model to the work type
Models are attached at the context profile level, so you can attach the model's personality to the type of work at hand.
Context Profiles bring the right sources, standards, skills, and review posture into each Thread, giving the workforce the focus each domain requires.
Models are attached at the context profile level, so you can attach the model's personality to the type of work at hand.
Switch out models instantly between providers any time at the context profile level.
Ask for the reviewing model to come from a different provider than the one that wrote the code, so the work is not checked by the same intelligence that produced it.
The approved list is an organization decision, and a project can narrow it further but never widen it. A profile naming a model outside that list does not get it.
The profile names the repos, docs, and knowledge a thread reads for this kind of work, which also means naming everything it should leave alone.
The capabilities this kind of work needs travel with the brief instead of being rediscovered, or missed, on every pass.
Every company has the same models available. What differs is the reasoning depth and the route this kind of work runs under, set once for everyone doing it inside the policy your organization already set.
What Lead Dev and your specialists know is a decision your team made and can read back, not a side effect of who happened to be at the keyboard.
Hospital discharge workflow
6 files
Waiting on the gate.
# Why the system is the way it is.
From day zero, Lead Dev reads the codebase and asks your team where the business is going, then turns both into knowledge every agent inherits. Lead Dev then continues to hone and craft this knowledge with you over time, delivering more valuable decisions for the business every day.
If it isn't in a format your agents can read, knowledge can't be used by your AI workforce. Codespeed centralizes your company's brain in a format Lead Dev and your specialist team can use from day zero to build value.
Every piece of knowledge is valuable to an agent, from business context to design thinking, brand guidelines, tone of voice, analytics, customer experience strategy, and everything in between. Give your agents the knowledge they need to understand who the customer is and what the business is trying to move.
Agentic skills were the early-stage solution to teaching agents. Codespeed's Knowledge Protocol extends far beyond these reusable primitives into highly specialized intelligence, carefully crafted and honed by your team of agents over time.
Just like your top engineering experts, Codespeed specialists focus on a precise domain deeply. Context profiles give those specialists the domain knowledge and skills they need to deliver their best work.
Turns well-scoped product needs into complete, tested changes across the stack.
Shipped the clinician scheduling flow behind a records integration.
Designs dependable APIs, services, data models, and migrations as the system grows.
Designed the claims resubmission path so attachments survive a failed transmission.
Builds polished interfaces and protects accessibility, interaction quality, and speed.
Rebuilt the dispatch console for crews working from a tablet in the field.
Improves delivery pipelines, environments, observability, and operational reliability.
Moved telemetry ingest to a second region without dropping the queue.
Keeps product intent, specifications, and team decisions aligned as the work moves.
Held the spec for a grid outage workflow while three teams changed it.
Finds friction, inconsistency, accessibility gaps, and unfinished customer journeys.
Walked a first-time analyst through the imagery review screen.
Keeps components, tokens, interaction patterns, and accessibility behavior consistent.
Kept one component set across a patient portal and its clinician console.
Studies onboarding, first value, conversion, and retention for high-leverage improvements.
Found where new labs stall between sample intake and first result.
Challenges implementation claims, checks risky assumptions, and returns evidence-backed verdicts.
Second-opinioned a rate limiter change before it reached a billing platform.
Builds focused coverage and verifies behavior across important happy and failure paths.
Proved a routing fix held across every carrier integration in the warehouse stack.
Checks release scope, migrations, rollback paths, and risk that is still unresolved.
Checked the migration and rollback path on a core banking release.
Reviews authentication, data exposure, dependencies, secrets, and sensitive product paths.
Reviewed ground station access before a partner network was added.
Finds latency, resource, and failure-mode issues before they turn into customer pain.
Chased the latency spike that appeared when a second plant came online.
Watches high-authority automation, production-sensitive paths, and unsafe action boundaries.
Bounded what automation may touch inside a regulatory reporting system.
Keeps guides, architecture notes, and onboarding material aligned with the product.
Kept the runbook for a 5G edge rollout current as it changed weekly.
Shapes clear product and customer-facing writing without losing the truth of the source.
Rewrote clinician-facing copy without losing the clinical meaning.
Designs repeatable agent evaluations and keeps claims tied to inspectable evidence.
Scored agent runs against a simulation code with known published results.
Finds instruction conflicts, context overload, missing grounding, and brittle behavior.
Tightened the context a specialist gets for a mainframe migration.
Reviews remembered guidance and recurring behavior for relevance, drift, and earned influence.
Corrected what the workforce learned from a bad claims release.
Studies model fit, latency, token use, and quality tradeoffs across real workloads.
Routed imagery pipeline work to the cheapest model that still passed.
Synthesizes primary sources, tests assumptions, and turns ambiguity into decision-ready findings.
Mapped how utilities handle outage communications across five regulators.
Tracks rival products, positioning, launches, and market shifts against customer value.
Tracked how rival carriers priced subscriber data plans through a launch.
Connects usage, billing, cost, and product behavior to customer and business outcomes.
Quantified what one failed resubmission costs an insurer per claim.
Routing isn't just a single decision at the start of a thread. It's intelligent adaptation during planning, execution, and review, within model boundaries your organization and context profile set.
Don't pay frontier prices for work that can be executed by cheaper models. Let Codespeed both orchestrate the work and intelligently route the right model to the right work, so you pay for what you need and not what you don't.
Codespeed knows the difference between a model saying the work is fine and a test proving it. Unmet requirements block the work from coming back, so cheaper routing never quietly becomes lower standards.
Codespeed is built on deep orchestration tools led by Lead Dev alongside context profile routing that keeps agents focused. That foundation ensures our model routing is highly intelligent and precise.