Talks & Workshops9 topics01EngineeringSoftware · Data · AI02ExecutivesAgentic AI · Productivity03AcademiaResearch · Method · Teaching45 min of content · 15 of debate

Research · Talks & Workshops

Talks & Workshops

Frontier topics in agentic AI, distributed systems, cloud and data — for engineers, executives and academia.

Carlos Diego C. P. · MIT Visiting Fellow · CESAR School
Founder, CEO & CTO at Valcann

9 topics
3 audiences
45 min + 15 of debate
PT · EN · ES

Invite me to speak →See the topics ↓

Frontier without evidence is just a trend. Every topic here comes from published research or a product in operation.

How the topics were chosen

01 · Audience

Software, Data and AI Engineers

For teams already putting agents and models into production — who need engineering, not magic.

Featured

1.1 · Featured topic

Agents are distributed systems — and almost nobody treats them that way

A multi-agent system inherits every classic distributed-systems problem — partial failure, shared state, retries that duplicate actions, ordering, consistency — with one aggravating factor: every node is non-deterministic. The talk applies to agents the toolkit engineering already masters (idempotency, sagas, circuit breakers, observability, SLOs and error budgets) and proposes what SRE for agents would look like: how to design, measure and operate systems that may answer the same question differently.

To open the debate

What is the SLO of a system that can give two different answers to the same input?

Ideal for: platform, SRE, architecture and AI engineering teams.

Multi-agent systems
SRE
Reliability

1.2

Capacity planning in the age of inference

The scarce resource is no longer CPU but GPU, tokens per second and context memory. Agentic workloads break any history-based model: a single task can trigger anywhere from 3 to 300 model calls. The talk extends C2PF to size inference workloads with no precedent — and replaces “how many machines?” with “what does it cost to complete a task, at what reliability?”.

To open the debate

Should cost per task be the primary SLO of an AI product?

Ideal for: cloud architects, FinOps and AI engineering leaders.

Inference
GPU
FinOps

1.3

When the specification becomes the source code

If the agent writes the code, the engineer’s artifact becomes the constraint: specification, guardrails and acceptance criteria. Using X-PRO.ai as a case, the talk shows spec-driven engineering in practice — the same agent behaves differently in a prototype and in a regulated system because it receives different contracts — and discusses which skills rise and which disappear.

To open the debate

Does reviewing agent-generated code still make sense, or should we review only the specification?

Ideal for: developers, tech leads and engineering managers.

Spec-driven
Coding agents
X-PRO.ai

02 · Audience

Business executives

Real-world applicability of AI and agentic AI, productivity gains that reach the bottom line, and strategy when intelligence becomes a cheap input.

Featured

2.1 · Featured topic

The agentic enterprise: when the org chart includes software

Automating isolated tasks yields little. The leap comes from redesigning processes on the assumption that part of the work is carried out by agents — with delegated authority, spending limits and human escalation. The talk presents a practical model to decide what to delegate, what to supervise and what remains exclusively human, and shows how to govern a workforce that mixes people and software.

To open the debate

Which decision in your company would you let an agent make on its own tomorrow?

Ideal for: CEOs, C-suite, boards and transformation leaders.

Agentic AI
Operating model
Governance

2.2

Why AI still doesn’t show up on your balance sheet

Pilots multiply, but results never reach the P&L. The talk explains the AI productivity paradox — individual gains are real, but they dissolve when the process doesn’t change — and shows where they do appear: with C2PF, research turned into process cut the proposal cycle by 87% across 93 engagements. Objective criteria to tell real gains from innovation theater.

To open the debate

Does your company measure AI by adoption or by results?

Ideal for: CFOs, COOs, operations and innovation directors.

Productivity
ROI
Process redesign

2.3

Strategy when thinking gets cheap

The cost of synthetic intelligence plunges with every model generation. If reasoning becomes a commodity, competitive advantage migrates to proprietary data, workflows, distribution and trust. Drawing on research into the economics of the generative AI value chain, the talk shows where value is concentrating — and which moats still hold.

To open the debate

If your competitor has access to the same model you do, what is your differentiator?

Ideal for: strategy teams, boards, investors and founders.

Strategy
AI economics
Competitive advantage

03 · Audience

Academics

Research agendas, method and teaching for a computing field in which the developer, the data and the compute have changed in nature.

Featured

3.1 · Featured topic

Empirical software engineering when the developer is an agent

Productivity, technical debt, quality, effort: the field’s core constructs were defined for humans. What happens to their validity — and to our research methods — when the author of the code is an agent? The talk examines what still measures what it claims to measure, what must be redefined, and proposes a research agenda for the next decade of software engineering.

To open the debate

Do the empirical studies we have still describe actual practice?

Ideal for: graduate programs, research groups and scientific conferences.

Empirical software engineering
Method
Research agenda

3.2

Systems that learn from what they themselves generated

The web is filling up with synthetic content, and the next models will be trained on it. Building on the Data-Centric Trust Pipeline, the talk discusses model collapse, the limits of synthetic augmentation — which did not reduce bias in the experiments — and provenance as a condition for verifiable data-driven research.

To open the debate

How do we ensure reproducibility when the data itself is generated by a model?

Ideal for: AI, data science and computing ethics researchers.

Synthetic data
Provenance
Reproducibility

3.3

Frontier science without a compute frontier

Computing power is concentrated in a few companies and a few countries. What does that mean for research done outside that axis — including in Brazil? The talk discusses which questions remain within academia’s reach and where it keeps an edge: evaluation, reliability, domain data and theory.

To open the debate

Should universities compete on scale or change the questions?

Ideal for: university leadership, research offices, funding agencies and faculty.

Science policy
Compute
Global South

Format

Asking the right questions matters as much as making the claims.

Every talk backs its claims with evidence — but doesn’t stop there. It ends with a hard question, framed with the same rigor as the answer, to open the debate rather than close it. At the frontier, the right question is worth more than a ready-made answer. All topics are tailored to the organization’s industry and moment.

Standard

Talk + debate

For events, conferences and internal meetings, with dedicated time for debate.

Main stage

Keynote

A condensed version for opening or closing events.

In-company

Workshop

Hands-on immersion with your team, using your organization’s own context and challenges.

The speaker

Research, industry and the classroom on the same stage

MITVisiting Researcher (Visiting Fellow), based in Boston, MA.

23 yearsworking in the technology industry.

16 yearsteaching and researching Computer Science.

›Founder, CEO & CTO at Valcann — AWS Collaboration Partner of the Year in Latin America (2024), part of EPI-USE / Group Elephant.
›Featured research in AI and Ethics (Springer, 2026) and at ICSE 2021; author of Cloud-Native Software Architecture.
›Recent stages: CIO Meeting, Gatua Summit Nordeste, B2B Tech Talk (Washington DC) and the Jon Myer Podcast (New York). See events →
›Professor at CESAR School, in the Computer Science undergraduate program, and in the Software Engineering master’s and doctoral programs.

Invitations & schedule

Bring one of these topics to your organization

Tell me about the audience, the format and the date. Talks in Portuguese, English and Spanish.

Get in touch →cdiego@cdiego.com

Corporate events · Conferences · Universities · In-company

← Ongoing research Contact →