STEP 1
Readiness assessment
Know where you stand.
Scores 6 areas with evidence. You leave with a scorecard, a prioritized fix list and a first-90-days plan.
See the Claude & Claude Code readiness assessment →01 // ARCHITECTConsulting & Architecture
Know what Claude does first. Then what earns the next wave.
Your readiness scorecard shows where you stand. The roadmap decides what Claude, Claude Code and the Claude API do first, what proves each step works, and what scales next, with owners, gates and a budget for every wave.
45 minutes. No pitch deck, no sales script.
- 01
Starts from your scorecard
Readiness findings become wave 0, not a separate report.
- 02
4 waves, 1 plan
Foundations, prove, scale and compound, each ending at a gate.
- 03
Build-ready first wave
Charters and a backlog your team can start the next day.
Where this fits
Readiness shows where you stand. The roadmap decides what happens next.
The roadmap is the second step of one path. It picks up where the Claude & Claude Code readiness assessment ends and turns its fix list and plan into a funded, sequenced program.
YOU ARE HERE
AI roadmap
Decide what happens next.
Turns the fix list into 4 waves with owners, gates, costs and a scored use case portfolio, then writes the backlog for wave 1.
STEP 3
Build and launch
Ship what the plan says.
Mach Engineering & Development builds wave 1 and GTM Enablement & Launch takes it to market. Or your team runs it.
From a score to a plan you can fund and build.
Assessed with someone else? Bring the scorecard. No scorecard yet? We run a short baseline across the same 6 areas first.
The problem: the planning gap
A scorecard is not a plan. The missing piece is sequence.
Readiness tells you what is in the way. Without a sequence, teams fix the loudest gap, start the pilot that demos best and lose the next quarter. 4 patterns that stall Claude programs after the assessment.
A fix list with no sequence
Every gap looks urgent. Without an order, the first pilot waits on a security review nobody scheduled.
Pilots with no exit rule
A pilot with no pass threshold cannot graduate. Nobody can say whether to scale, hold or stop.
Foundations built per team
Each team writes its own CLAUDE.md, connectors and prompts. Spend climbs and nothing is reusable.
Usage nobody can measure
Seats and sessions rise. Nobody can tie them to cycle time, quality or cost per task.
From scorecard to roadmap
6 readiness areas become 6 workstreams. Nothing is scored and then dropped.
Every gap in your scorecard becomes a funded line in the roadmap, with an owner and a wave.
01Data readiness
Data and access
Which connectors and MCP servers are approved, in the order the use cases need them, with access scoped by role.
02Use case clarity
Portfolio
Every use case scored on value and effort, with a named owner and a baseline, and re-scored every quarter.
03Platform review
Platform
Claude Code environments, CI/CD and integration work sequenced so shared foundations are built once.
04Security and governance
Controls
Managed settings, permissions, review rules and human approval points set before the work that needs them.
05Evaluation maturity
Evidence
A first eval set from real tasks, pass thresholds for each wave and cost per task tracked from wave 1.
06Team and skills
Enablement
Champions, role playbooks and training plans, timed to each rollout.
Where you start
Your score sets the starting wave.
The self-check on the readiness page is directional. The assessment scorecard replaces it with evidence.
Start in wave 0
Close the access, security and evaluation gaps first. One low-risk pilot runs beside them so the team keeps learning.
Start in wave 1
Run 2 scoped pilots behind eval gates while the foundations finish in parallel.
Start in wave 2
Extend what works: shared tools and repo standards, automation goals and a quarterly re-score.
The 5 layers
The same 5 layers. Now with an order.
A wrong call in one layer forces rework in the rest. The roadmap sequences all 5 together, so wave 1 does not depend on a decision nobody has made.
01
Claude apps
Team and Enterprise plansProjectsConnectorsSSO and provisioningAdmin and usage controls
First move
2 role playbooks with named champions, on approved connectors only.
Then
More roles and a connector catalog in wave 2. Shared Projects for repeat work in wave 3.
02
Claude Code
Managed settingsCLAUDE.md standardsHooks and skillsSubagent reviewCI/CD and GitHub
First move
Managed settings and a permission baseline, then 1 repo with CLAUDE.md, tests as the check and hooks for rules that must always run.
Then
More repos on the same standards, shared skills and subagents, and non-interactive runs in CI where a check exists.
03
Claude API and Agent SDK
Workflows before agentsTool designMCP serversEval gatesPrompt caching and batch
First move
1 workflow built with the simplest pattern that passes its evals.
Then
A shared tool and MCP library, caching and batch for cost, and more autonomy only where evals justify it.
04
Cloud and procurement
Anthropic APIAmazon BedrockGoogle Cloud Vertex AIMicrosoft Foundry
First move
A cloud path decision record before anything is built: region, residency, identity and who pays.
Then
Commit sizing from real usage in wave 2.
05
Governance and people
Data handling and DLPAudit and loggingModel risk reviewTraining and championsSuccess metrics
First move
Policy, review owners and human approval points for high-risk actions.
Then
A champion network and training in wave 2, and a quarterly governance review from wave 3.
The roadmap
4 waves. Each one earns the next.
Waves overlap. Each one ends at a gate with 3 possible decisions: go, hold or redesign. The horizon is sized to your organization and the whole plan is re-scored every quarter.
WAVE 0Starts in Q1
Foundations
Make it safe to start.
- Cloud path decision record (Anthropic API, Bedrock, Vertex AI or Foundry)
- Managed settings and a Claude Code security baseline
- Approved connector and MCP server list, with data access scoped by role
- First eval set of 20 to 50 real tasks
Owners Security lead, platform lead, program owner
Gate
Security sign-off. Eval baseline recorded. Every use case has a named owner.
WAVE 1Q1 to Q2
Prove
Earn the right to scale.
- 2 pilots, each with an owner, a baseline and an exit rule
- Claude apps playbooks for 2 roles, with champions
- Claude Code in 1 repo: CLAUDE.md, tests as the check, fresh-context review
- 1 API workflow, built with the simplest pattern that passes its evals
Owners Use case owners, engineering lead, champions
Gate
Eval thresholds met. Cost per task known. Use beyond the first team. Go, hold or redesign.
WAVE 2Q2 to Q4
Scale
Repeat what worked.
- More teams and repos on the same standards
- A shared library of skills, subagents, tools and MCP servers
- Claude Code in CI through non-interactive runs, where a check exists
- Commit sizing from real usage, a champion network and training
Owners Platform lead, team leads, finance partner
Gate
Regression evals pass on every change. Spend inside budget. Impact measured by outcome, not seats.
WAVE 3Q3 to Q4
Compound
Automate more. Re-score every quarter.
- Raise autonomy only where evals justify it
- Move capability evals that saturate into regression suites
- Re-score the portfolio and re-order the waves each quarter
- Quarterly governance and model risk review
Owners Program owner, governance board
Gate
Quarterly refresh signed off and funded.
The graphic above is illustrative. Timing is sized to your organization and the whole plan is re-scored every quarter.
Built on public guidance
The plan follows what Anthropic publishes.
We plan against Anthropic's public guidance on agents, Claude Code, evaluations and AI fluency, and we link each source so your team can read it. Mach is an independent firm. Linking to these documents is not an endorsement.
Start with the simplest solution
Wave 1 uses a single call or a fixed workflow before any agent. Autonomy is a wave 3 decision, taken only where evals show it helps.
Give Claude a check it can run
Every Claude Code use case enters wave 1 with tests or another pass or fail check. If we cannot define one, it waits.
Build evals from real tasks
Wave 0 seeds 20 to 50 tasks from real work. Capability evals that saturate become regression suites by wave 3.
Use hooks for rules that cannot be optional
Managed settings and hooks carry the rules that must run every time. CLAUDE.md carries guidance and stays short.
Review in a fresh context
Changes that touch production get a second session or a subagent review, plus a human for high-risk work.
Build fluency, not only access
Training plans follow the 4 competencies of Anthropic's AI Fluency framework: delegation, description, discernment and diligence.
SourceAI Fluency framework ↗
Deliverables
What you walk away with.
Everything is yours to keep, whether you build with us or on your own.
01PDF
Claude roadmap on a page
4 waves with owners, gates, costs and success measures, one page per wave.
02Sheet
Scored use case portfolio
Every Claude use case ranked on value and effort, with a named owner and a baseline.
03Diagram
Target-state Claude architecture
Which surface serves which use case, how data reaches it and where the controls sit.
04Plan
Eval plan and stage gates
The first eval tasks, pass thresholds and the go, hold or redesign rule for each wave.
05Checklist
Claude Code and governance model
Managed settings, repo standards, review rules and human approval points, each with an owner.
06Backlog
Wave 1 charters and backlog
2 pilot charters and build-ready epics your team or ours can start the next day.
How it works
4 steps, one team.
Senior operators work alongside your team, so every wave is backed by evidence, not opinion.
STEP 01
Anchor
Start from your readiness scorecard and fix list. No scorecard yet? We run a short baseline across the same 6 areas first.
STEP 02
Prioritize
Collect use cases from every team, score them on value and effort, and agree what goes first and why.
STEP 03
Sequence
Build the 4 waves with dependencies, owners, costs and gates, and write the eval plan that each gate uses.
STEP 04
Hand off
Run the executive session, then turn wave 1 into charters and a backlog your team can start on.
Why Mach Advisory™
The roadmap and the build come from one team.
A Claude roadmap written without knowing how Claude Code runs in your repositories, or how connectors reach your data, is a slide. Mach carries the same plan from assessment through architecture, build and launch, so wave 1 is a backlog and not a recommendation.
You can take the roadmap and build it yourself. Many clients use us for some waves and their own engineers for others.
What does Claude do first, and what has to be true before it does more?
The question this roadmap answers
Powered by our platform CloudTeams™
Context stays on one record, not in a deck
The Claude roadmap runs on the CloudTeams™ Platform, the AI-powered platform we built for firms that sell expertise. Mach is its first customer.
- Contacts, notes and tasks from discovery stay on one record
- A read-only AI assistant answers questions across it, so nothing is re-explained
- One team carries the work from readout to roadmap
FAQ
Questions before the roadmap
Straight answers. Still curious? Ask us on a free call.
How is the roadmap different from the readiness assessment?
The assessment scores 6 areas and shows what is in the way. The roadmap decides what to do about it: which use cases go first, in what order, with which owners, gates and budget. The assessment ends with a first-90-days plan. The roadmap extends it into 4 waves and a build-ready backlog.
Do we need the readiness assessment first?
No, but it makes the roadmap faster and better grounded. If you have a scorecard, we start from it. If you do not, we run a short baseline across the same 6 areas before we sequence anything.
Which parts of Claude does the roadmap cover?
Claude apps, Claude Code, and the Claude API and Agent SDK, plus the cloud path you buy through. Each wave names which surface serves which use case, so the plan does not treat them as one product.
Is this a vendor evaluation?
No. We plan from your organization, data and platforms to what your Claude plans require. The roadmap does not compare vendors. If you also use ChatGPT or Codex, there is a matching roadmap for those, and the two can share one portfolio and one set of gates.
Where do the recommendations come from?
From your evidence and from Anthropic's public guidance on agents, Claude Code and evaluations, which we link on this page. Mach is an independent firm, and linking to those documents is not an endorsement by Anthropic.
How long does it take?
It depends on the number of teams, systems and use cases in scope. We confirm the timeline on the first call. Wave 1 is always defined in enough detail to start building.
How often should the roadmap change?
Review it every quarter. Use cases that are high effort today get easier as models and tools improve, so the order should move. Each wave ends at a gate where you decide to go, hold or redesign.
Can Mach build wave 1?
Yes. Our Engineering & Development practice can build it and our GTM practice can launch it, with the same team. You can also take the backlog to your own engineers or another partner.
Who owns what you produce?
You do. Everything is yours to keep, whether you build with us or on your own.
Claude and Claude Code are products of Anthropic, PBC. The Anthropic logo is a trademark of Anthropic, PBC. Mach Advisory™ is an independent firm. Guidance cited on this page comes from Anthropic's public documentation, and linking to it does not imply endorsement. Product capabilities and availability change frequently, so we confirm current details with you at scoping. The roadmap shown is illustrative and is not client data.
Project inquiry
Tell us about your project.
Tell us where you are and what you need. We will tell you honestly whether a Claude roadmap is the right next step.
- 01Tell us where you are headedA two-minute form. Share the outcome you want and any constraints.
- 0245-minute discovery callWe listen first. No pitch deck, no sales script.
- 03Your readout within one weekWhere the gaps are and which practice will move the needle fastest.
First conversation freeSenior team, not a sales repMutual NDA on request
Ready to choose a time? Schedule your free consultation →
Project inquiry // 2 minutes
Tell us about a Claude roadmap
Sending this form requests a follow-up; it does not reserve a meeting time. Prefer a new tab? Open the inquiry form